Odia STT v0.1.0

Odia STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Odia / ଓଡ଼ିଆ / or
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{odia-stt, author = {Tyers,Francis}, title = {Odia STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-OR-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Odia Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	98.9%	55.2%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Mongolian STT v0.1.0

Mongolian STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Mongolian / Монгол хэл / mn
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{mongolian-stt, author = {Tyers,Francis}, title = {Mongolian STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-MN-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Mongolian Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	96.7%	45.5%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Maltese STT v0.1.0

Maltese STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Maltese / Malti / mt
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{maltese-stt, author = {Tyers,Francis}, title = {Maltese STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-MT-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Maltese Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	93.6%	33.7%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Luganda STT v0.1.0

Luganda STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Luganda / Lugdanda / lg
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{luganda-stt, author = {Tyers,Francis}, title = {Luganda STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-LG-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Luganda Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	97.7%	33.2%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Lithuanian STT v0.1.0

Lithuanian STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Lithuanian / Lietuvių kalba / lt
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{lithuanian-stt, author = {Tyers,Francis}, title = {Lithuanian STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-LT-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Lithuanian Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	98.9%	36.0%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Latvian STT v0.1.0

Latvian STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Latvian / Latviešu valoda / lv
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{latvian-stt, author = {Tyers,Francis}, title = {Latvian STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-LV-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Latvian Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	88.3%	31.1%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Kyrgyz STT v0.1.0

Kyrgyz STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Kyrgyz / Кыргызча / ky
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{kyrgyz-stt, author = {Tyers,Francis}, title = {Kyrgyz STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-KY-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Kyrgyz Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	94.1%	36.8%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Irish STT v0.1.0

Irish STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Irish / Gaeilge / ga-IE
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{irish-stt, author = {Tyers,Francis}, title = {Irish STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-GA_IE-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Irish Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	94.3%	57.7%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Indonesian STT v0.1.0

Indonesian STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Indonesian / Bahasa indonesia / id
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{indonesian-stt, author = {Tyers,Francis}, title = {Indonesian STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-ID-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Indonesian Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	89.7%	30.3%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.

Hungarian STT v0.1.0

Hungarian STT v0.1.0 (ITML)

Jump to section:

• Model details
• Intended use
• Performance Factors
• Metrics
• Training data
• Evaluation data
• Ethical considerations
• Caveats and recommendations

Model details

• Person or organization developing model: Originally trained by Francis Tyers and the Inclusive Technology for Marginalised Languages group.
• Model language: Hungarian / Magyar nyelv / hu
• Model date: April 9, 2021
• Model type: Speech-to-Text
• Model version: v0.1.0
• Compatible with 🐸 STT version: v0.9.3
• License: AGPL
• Citation details: @techreport{hungarian-stt, author = {Tyers,Francis}, title = {Hungarian STT 0.1}, institution = {Coqui}, address = {\url{https://github.com/coqui-ai/STT-models}} year = {2021}, month = {April}, number = {STT-CV6.1-HU-0.1} }
• Where to send questions or comments about the model: You can leave an issue on STT-model issues, open a new discussion on STT-model discussions, or chat with us on Gitter.

Intended use

Speech-to-Text for the Hungarian Language on 16kHz, mono-channel audio.

Performance Factors

Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors here.

Metrics

STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.

Transcription Accuracy

The following Word Error Rates and Character Error Rates are reported on omnilingo.

Test Corpus	WER	CER
Common Voice	89.2%	32.7%

Real-Time Factor

Real-Time Factor (RTF) is defined as proccesing-time / length-of-audio. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.

Recorded average RTF on laptop CPU: ``

Model Size

model.pbmm: 181M
model.tflite: 46M

Approaches to uncertainty and variability

Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.

Training data

This model was trained on Common Voice 6.1 train.

Evaluation data

The Model was evaluated on Common Voice 6.1 test.

Ethical considerations

Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.

Demographic Bias

You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.

Surveillance

Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.

Caveats and recommendations

Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data here.

In most applications, it is recommended that you train your own language model to improve transcription accuracy on your speech data.