7 ms·

My quick test of DeepSpeech, as posted on Facebook back in April: -- I just ran a speech-to-text converter on a very clear clip of former Doctor Who actor Tom

My quick test of DeepSpeech, as posted on Facebook back in April: -- I just ran a speech-to-text converter on a very clear clip of former Doctor Who actor Tom Baker talking in an interview. The DeepSpeech converter uses the very latest AI deep-learning advancements to 'listen' to the audio and output the spoken words as text. After 3 long minutes of running it on a 30-second clip, it printed out its interpretation: "hooloomooloo how booboorowie i have a honeymoon"

fxtentacle 6y ago
When I tried it out with English and German phone recordings, it was working competitively. I would have ranked it better than Amazon but worse than Google. Did you maybe not convert your WAV to the correct sampling rate?
donw 6y ago
I didn’t know Tom Baker was Welsh.
twoslide 6y ago
It's supposed to work on sentence long audio (4 - 5 seconds), they suggest chunking your audio first: https://discourse.mozilla.org/t/longer-audio-files-with-deep-speech/22784 https://discourse.mozilla.org/t/longer-audio-files-with-deep...
- magicalhippo 6y ago
  Also there's essentially two parts to this, the neural net is used for speech-to-characters, and then a language model is used to convert the character stream to words. I found that the language model they supplied was trained data that did not contain the words I needed, and got significantly improved results when making my own language model using the kenlm[1] tools. [1]: https://kheafield.com/code/kenlm/ https://kheafield.com/code/kenlm/
  - zaptrem 6y ago
    Would it be possible to substitute this for GPT2/BERT? Or is that a different type of language model? Can the pre-trained language model be fine-tuned? I’m using DeepSpeech to transcribe long-form lecture audio, and have just assumed there would be a massive improvement once they noise-harden the models with 1.0.
    - nshm 6y ago
      GPT2 is not a good language model but there are things like XLM. Mozilla DeepSpeech doesn't support XLM rescoring, other toolkits do and it gives great improvement in accuracy. If you care about accurate transcription you'd better consider alternatives.
      - zaptrem 6y ago
        I didn't know any other ML-based open source transcription engines existed? I can't seem to find them on Google.
bmn__ 6y ago
Experiment is worthless for drawing conclusions if it's not reproducible by other people. Besides the hypothesis that DS sucks, the software could also very well be just fine and you made methodological errors.

fxtentacle 6y ago

When I tried it out with English and German phone recordings, it was working competitively. I would have ranked it better than Amazon but worse than Google. Did you maybe not convert your WAV to the correct sampling rate?

donw 6y ago

I didn’t know Tom Baker was Welsh.

twoslide 6y ago

It's supposed to work on sentence long audio (4 - 5 seconds), they suggest chunking your audio first: https://discourse.mozilla.org/t/longer-audio-files-with-deep-speech/22784 https://discourse.mozilla.org/t/longer-audio-files-with-deep...

magicalhippo 6y ago
Also there's essentially two parts to this, the neural net is used for speech-to-characters, and then a language model is used to convert the character stream to words. I found that the language model they supplied was trained data that did not contain the words I needed, and got significantly improved results when making my own language model using the kenlm[1] tools. [1]: https://kheafield.com/code/kenlm/ https://kheafield.com/code/kenlm/
- zaptrem 6y ago
  Would it be possible to substitute this for GPT2/BERT? Or is that a different type of language model? Can the pre-trained language model be fine-tuned? I’m using DeepSpeech to transcribe long-form lecture audio, and have just assumed there would be a massive improvement once they noise-harden the models with 1.0.
  - nshm 6y ago
    GPT2 is not a good language model but there are things like XLM. Mozilla DeepSpeech doesn't support XLM rescoring, other toolkits do and it gives great improvement in accuracy. If you care about accurate transcription you'd better consider alternatives.
    - zaptrem 6y ago
      I didn't know any other ML-based open source transcription engines existed? I can't seem to find them on Google.

bmn__ 6y ago

Experiment is worthless for drawing conclusions if it's not reproducible by other people. Besides the hypothesis that DS sucks, the software could also very well be just fine and you made methodological errors.