← Back to Showcase

Project collection · Ongoing

SpeeD-TB

Speech Datasets and Models for Tibeto-Burman Languages of India

Dataset created under the Speech Datasets and Models for Tibeto-Burman Languages (SpeeD-TB), sponsored under Mission Bhashini by Ministry of Electronics and Information Technology (MEITY), Govt of India. The project aimed to create 1,200 hours of speech dataset and ASR models for 6 underresourced, tribal Tibeto-Burman languages of India speoken in Eastern and North-Eastern parts of India.

Explore the language map →
1,476.7
Hours
6
Languages
345465
Entries
2066
Speakers
Project networkLeadership, partners and funders
24 records

dataset

SpeeD-TB Nyishi Narration (Phase 1)

About Nyishi

Nyishi is an under-resourced Tibeto-Burman language spoken by the Nyishi people, the largest ethnic group in Arunachal Pradesh, India. According to the 2011 Census, there are approximately 3 lakh speakers of the language. The language belongs to the Tani branch of the Tibeto-Burman language family. Owing to its status as one of the largest languages of Arunachal Pradesh, some resources for language development, such as a corpus and a language model, have been developed. However, the language lacks any large, significant speech or text corpus.

Dataset Description

The Nyishi Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB) funded under Mission Bhashini, is a transcribed speech corpus of Nyishi. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it the ** largest speech resource for the language** that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Nyishi language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the full dataset for the language collected till now.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Toto Narration (Phase 1)

About Toto

Toto is an under-resourced and critically endangered Tibeto-Burman language spoken by a small community in Totopara village, Alipurduar district, West Bengal, India, with less than 1,000 speakers (and significantly lesser number of people proficient in the language). The language belongs to Dhimalish group of languages and is closely related to Dhimal, another language spoken in Northern West Bengal.

Dataset Description

The Toto Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB) funded under Mission Bhashini, is a transcribed speech corpus of Toto,. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Bengali script, making it the first and largest resource for the language that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Toto language by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. Rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby, giving a large coverage. We have also used a variety of elicitation methods for collecting the data including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to a rich metadats including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby, ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the full dataset for the language collected till now.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Kokborok Narration (Phase 1)

About Kok Borok

Kok Borok (ISO 639-3: trp, Glottocode: kokb1239), also known as Kokborok or Tripuri, belongs to the Tibeto-Burman language family. Within this family, it is classified under the Bodo-Garo branch of the Sal subfamily. According to the Census of India 2011, there are approximately 1,011,294 speakers of "Tripuri" (which includes Kok Borok as the dominant variety) in India. The principal region of Kok Borok speakers is the state of Tripura in Northeast India, where it holds official status. Smaller communities of speakers are located in the neighbouring states of Assam and Mizoram, as well as in adjacent border areas of Bangladesh.

Dataset Description

The Kok Borok Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB), funded under Mission Bhashini, is a transcribed speech corpus of the language. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it ** one of the largest speech resources for the language**, which not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. All the questionnaires and datasets are being released as part of the project.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Meitei Narration (Phase 1)

Meitei (also known as Manipuri and Meetei; ISO 639-3: mni, Glottocode: meit1246 / mani1292) belongs to the Tibeto-Burman language family. Its precise subgrouping within Tibeto-Burman remains a subject of academic debate, often classified within its own independent branch or grouped tentatively with the Kuki-Chin-Naga languages. According to the 2011 Census of India, there are approximately 1.76 million native speakers of Manipuri in India. The language is primarily spoken in the northeastern state of Manipur, where it serves as the official state language and the lingua franca among diverse ethnic groups. Significant speaker communities also exist in the neighbouring states of Assam, Tripura, and Nagaland. Owing to its official status as one of the scheduled languages of India, a large amount of resource development work has been undertaken for the language. Our current corpus adds to the ever-increasing corpora being collected for the language.

Dataset Description

The Meitei Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB) funded under Mission Bhashini, is a transcribed speech corpus of the language. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it ** one of the largest speech resources for the language**, which not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. All the questionnaires and datasets are being released as part of the project.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Bodo Narration (Phase 1)

About Bodo

Bodo belongs to the Tibeto-Burman language family, specifically belonging to the Bodo-Garo subgroup of the Sal language group. According to the official 2011 Census of India, there are approximately 1.48 million native Bodo speakers. The principal concentration of Bodo speakers is in the state of Assam, particularly within the autonomous Bodoland Territorial Region. Smaller speech communities are also distributed across adjacent districts in the states of West Bengal, Meghalaya, and Nagaland.

Dataset Description

The Bodo Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB), funded under Mission Bhashini, is a transcribed speech corpus of the language. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it ** one of the largest speech resources for the language**, which not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. All the questionnaires and datasets are being released as part of the project.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Toto Pilot Narration (Lifecycle and Agriculture)

About Toto

Toto is an under-resourced and critically endangered Tibeto-Burman language spoken by a small community in Totopara village, Alipurduar district, West Bengal, India, with ** fewer than 1,000 speakers** (and a significantly smaller number of people proficient in the language). The language belongs to the Dhimalish group of languages and is closely related to Dhimal, another language spoken in Northern West Bengal.

Dataset Description

The Toto Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB) funded under Mission Bhashini, is a transcribed speech corpus of Toto,. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Bengali script, making it the first and largest resource for the language that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Toto language by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. Rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby, giving a large coverage. We have also used a variety of elicitation methods for collecting the data including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to a rich metadats including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby, ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This projects contains the data collected during the pilot phase of the project and contains narrations from the domains of lifecycle and agriculture

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Chokri Youtube (Phase 1)

Chokri (also known as Chakru, ISO 639-3: nri, Glottocode: chok1243) belongs to the Tibeto-Burman language family, specifically classified within the Angami-Pochuri branch of the Southern Tibeto-Burman subgroup.

Speakers

According to the 2011 Census of India, there are approximately 91,257 speakers of Chakru/Chokri in India.

Distribution in India

Chokri is predominantly spoken in the northeastern state of Nagaland, India. It is concentrated heavily in the Phek district, particularly across the Pfütsero, Chetheba, and Chazouba administrative circles.

Major grammatical features

  • Phonology: Chokri is a tonal language characterised by a complex pitch/tone system that distinguishes lexical meaning. Its consonant inventory is notable for contrasting aspirated and unaspirated stops, alongside a series of voiceless sonorants (including voiceless nasals and laterals) typical of Angami-Pochuri languages.
  • Morphology: The language is predominantly agglutinative. Morphological processes rely heavily on suffixation and prefixation for word formation. Nouns take possessive prefixes corresponding to person and number, and verbs are modified by a range of aspectual, modal, and directional markers.
  • Syntax: Chokri exhibits a basic Subject-Object-Verb (SOV) constituent word order. It is a postpositional language where modifiers like numerals and demonstratives generally follow the head noun, while relative clauses can precede or follow the noun they modify.

Dataset Description

The Chokri Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB) funded under Mission Bhashini, is a transcribed speech corpus of Chokri. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it the first and largest resource for the language that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Toto language by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. Rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby, giving a large coverage. We have also used a variety of elicitation methods for collecting the data including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to a rich metadats including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby, ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the data collected from the field using narration methods.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Chokri Narration (Phase 1)

Chokri (also known as Chakru, ISO 639-3: nri, Glottocode: chok1243) belongs to the Tibeto-Burman language family, specifically classified within the Angami-Pochuri branch of the Southern Tibeto-Burman subgroup.

Speakers

According to the 2011 Census of India, there are approximately 91,257 speakers of Chakru/Chokri in India.

Distribution in India

Chokri is predominantly spoken in the northeastern state of Nagaland, India. It is concentrated heavily in the Phek district, particularly across the Pfütsero, Chetheba, and Chazouba administrative circles.

Major grammatical features

  • Phonology: Chokri is a tonal language characterised by a complex pitch/tone system that distinguishes lexical meaning. Its consonant inventory is notable for contrasting aspirated and unaspirated stops, alongside a series of voiceless sonorants (including voiceless nasals and laterals) typical of Angami-Pochuri languages.
  • Morphology: The language is predominantly agglutinative. Morphological processes rely heavily on suffixation and prefixation for word formation. Nouns take possessive prefixes corresponding to person and number, and verbs are modified by a range of aspectual, modal, and directional markers.
  • Syntax: Chokri exhibits a basic Subject-Object-Verb (SOV) constituent word order. It is a postpositional language where modifiers like numerals and demonstratives generally follow the head noun, while relative clauses can precede or follow the noun they modify.

Dataset Description

The Chokri Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB) funded under Mission Bhashini, is a transcribed speech corpus of Chokri. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it the first and largest resource for the language that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Toto language by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. Rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby, giving a large coverage. We have also used a variety of elicitation methods for collecting the data including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to a rich metadats including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby, ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the data collected from the field using narration methods.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Nyishi Interviews and Conversations (Phase 1)

About Nyishi

Nyishi is an under-resourced Tibeto-Burman language spoken by the Nyishi people, the largest ethnic group in Arunachal Pradesh, India. According to Census 2011, there are approximately 3 lakh speakers of the language. The language belongs to the Tani branch of the Tibeto-Burman language family. Owing to its status as one of the largest languages of Arunachal Pradesh, some resources for language development, such as a corpus and a language model, have been developed. However, the language lacks any large, significant speech or text corpus.

Dataset Description

The Nyishi Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB), funded under Mission Bhashini, is a transcribed speech corpus of Nyishi. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it the ** largest speech resource for the language** that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Nyishi language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the full dataset for the language collected till now in the field using structured and semi-structured interviews and spontaneous conversation.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Nyishi Lectures (Phase 1)

About Nyishi

Nyishi is an under-resourced Tibeto-Burman language spoken by the Nyishi people, the largest ethnic group in Arunachal Pradesh, India. According to Census 2011, there are approximately 3 lakh speakers of the language. The language belongs to the Tani branch of the Tibeto-Burman language family. Owing to its status as one of the largest languages of Arunachal Pradesh, some resources for language development, such as a corpus and a language model, have been developed. However, the language lacks any large, significant speech or text corpus.

Dataset Description

The Nyishi Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB), funded under Mission Bhashini, is a transcribed speech corpus of Nyishi. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it the ** largest speech resource for the language** that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Nyishi language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the full dataset for the language collected till now using long-form lectures (approx. 30 minutes for each lecture).

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Nyishi Translations (Phase 1)

About Nyishi

Nyishi is an under-resourced Tibeto-Burman language spoken by the Nyishi people, the largest ethnic group in Arunachal Pradesh, India. According to Census 2011, there are approximately 3 lakh speakers of the language. The language belongs to the Tani branch of the Tibeto-Burman language family. Owing to its status as one of the largest languages of Arunachal Pradesh, some resources for language development, such as a corpus and a language model, have been developed. However, the language lacks any large, significant speech or text corpus.

Dataset Description

The Nyishi Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB), funded under Mission Bhashini, is a transcribed speech corpus of Nyishi. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it the ** largest speech resource for the language** that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Nyishi language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the full dataset for the language collected so far using the translation method. The sentences are translated from English to Nyishi, thereby producing a parallel corpus of the language.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →

dataset

SpeeD-TB Nyishi Youtube (Phase 1)

About Nyishi

Nyishi is an under-resourced Tibeto-Burman language spoken by the Nyishi people, the largest ethnic group in Arunachal Pradesh, India. According to Census 2011, there are approximately 3 lakh speakers of the language. The language belongs to the Tani branch of the Tibeto-Burman language family. Owing to its status as one of the largest languages of Arunachal Pradesh, some resources for language development, such as a corpus and a language model, have been developed. However, the language lacks any large, significant speech or text corpus.

Dataset Description

The Nyishi Speech Dataset, developed as part of the Speech Datasets and Models for Tibeto-Burman Languages (Project SpeeD-TB), funded under Mission Bhashini, is a transcribed speech corpus of Nyishi. The full dataset comprises over 200 hours of high-quality audio recordings paired with accurate transcriptions in both IPA and Roman script, making it the ** largest speech resource for the language** that not only enables building and evaluating voice models in low-resource linguistic settings but also enables large-scale linguistic description and documentation of the language. The audio data captures a diverse range of speakers across different age groups, genders, and education levels, ensuring variability in pronunciation, speech patterns, and tone. It includes both spontaneous and read speech collected in naturalistic and semi-controlled environments, thereby reflecting real-world linguistic usage. The transcriptions are carefully prepared and normalised to maintain consistency, supporting robust model training. This dataset also contributes to the preservation and digital documentation of the Nyishi language and culture by transforming oral knowledge into structured, machine-readable formats.

Almost 60% of the data in the corpus is included from domains of agriculture, education and science & technology. The rest of the data is from varied domains including culture, lifecycle, sports, entertainment, healthcare and oral history, thereby giving extensive coverage. We have also used a variety of elicitation methods for collecting the data, including translations, narrations, lectures, role-play, spontaneous conversations, interviews and picture and video descriptions. The released dataset is meticulously mapped to rich metadata, including demographic and linguistic metadata of the speakers, domains, elicitation methods and to individual prompts. The audio included in the current dataset is already sliced at sentence level, thereby ready to be integrated into the model training pipeline out-of-the-box.

The overall dataset of the project is collected over multiple phases and using multiple questionnaires. This repository contains the full dataset for the language collected till now from YouTube sources.

Ethical Considerations, IPR and Attribution

This repository represents our commitment to not only fair remuneration to the speakers of the language but also to co-ownership and equal IPR to all the contributors who have built the dataset. We believe this is the first step to move away from extractive data collection and use practices and ensure fairness in our treatment of the community members. As such, we have listed all speakers and transcribers as Contributors to the dataset. This dataset is only licensed to other researchers for use in their research projects. More details about licensing and commercial use conditions are given in the License and Commercial Use sections.

full CC BY-NC-SA 4.0 Open resource → Explore relationships →