DeepSpeech

One other "AI" thing that I really liked came before the LLM boom. Mozilla made a speech-to-text engine called [DeepSpeech](https://github.com/mozilla/DeepSpeech) for running on-device when you use dictation in the browser. Apparently, it was discontinued last year, so I'm not sure what they use now.

As I understand it DeepSpeech is no longer actively maintained by Mozilla: https://github.com/mozilla/DeepSpeech/issues/3693 For Text To Speech, I've found Piper TTS useful (for situations where "quality"=="realistic"/"natual"): https://github.com/rhasspy/piper For Speech to Text (which AIUI DeepSpeech provided), I've had some success with Vosk: https://github.com/alphacep/vosk-api

DeepSpeech? https://github.com/mozilla/DeepSpeech

https://github.com/mozilla/DeepSpeech maybe

the technology you're looking for is "on-device" or "local" speech-to-text https://cloud.google.com/blog/products/ai-machine-learning/s... https://picovoice.ai/platform/cheetah/ https://github.com/mozilla/DeepSpeech

Mozilla did release DeepSpeech[0] and Firefox Translation[1] (the latter of which they included in Firefox, to offer client-side webpage translations.) They definitely have fewer resources than OpenAI, and they do not produce SOTA research (their publications have plummeted to 1/year anyway[2]). So the only way for them to make progress is to seek government grants or make challenges like these. This challenge is unlikely to be profitable for the winning team: the expected value of winnings are likely around $1K when taking into account the probability that another team gets a better rank, but ML research projects are often more expensive (recently, Alpaca spent upwards of $600 on computation alone; and of course pretraining large models is much more expensive). So the main gain will be publicity. [0]: https://github.com/mozilla/deepspeec

Mozilla has an open source tect-to-speech project, called deep speech. https://github.com/mozilla/DeepSpeech

> mozillas common voice. How can it be used for transcription? In their website I only see an interface for either uploading audio or submitting transcriptions: https://commonvoice.mozilla.org/es The Github repo they mention (https://github.com/common-voice/common-voice) seems to be just that sample collection software. I do not see where I can download the software to transcribe audio. Edit: All right, the speech recognition part is actually called DeepSpeech: https://opensource.com/article/22/1/voice-text-mozilla-deeps... https://github.com/mozilla/DeepSpeech

In my opinion, the focus here should be on the fact that this is a state of the art AI which beats Facebook's wav2vec2 by a relative 16% improvement, Scribosermo (based on NVIDIA NeMo) by a relative 44% and Mozilla's DeepSpeech German by a relative 75% improvement. People usually don't share their production-quality tools ;) That said, I wrote "284 lines of C++" to indicate that this is compact enough for people to actually read and understand the source code. Also, compiling my implementation is super easy and straightforward ... something which can't be said for Kaldi, Vosk, or DeepSpeech. If you try to read the CTC beam search decoder from Mozilla's DeepSpeech [1], that alone is about 2000 LOC in multiple files. If you try to read the pyctcdecode source that is used by HuggingFace [2], that's 1000+ LOC of Python. But this implementatio

IIRC every browser that supports the Web Speech API does so via cloud services. Mozilla being the only major browser engine maker without it's own "cloud" and having slightly fewer phone-home features than many of the others didn't want to do that. Mozilla has been doing quite a bit of work in the area though (for example https://github.com/mozilla/DeepSpeech), hopefully to enable these features locally in the future. IMO it's good if we don't make web platform features unfeasible to implement locally, so I think mozillas stance makes sense. A new browser should not have to have the backing of a massive cloud provider willing to give away compute for free.

This is more of a "DIY" approach but all the tools are there for FOSS and OSHWA solutions. Mozilla has DeepSpeech [0] and, while not as advanced as the stuff from Google or Amazon, my experimentation left me feeling pretty hopeful that it could reliably recognize at least keywords. The Raspberry Pi is quite capable though you'll probably need some dedicated microphone to reliably catch voice data. I know ReSpeaker [1] but maybe some off the shelf conference USB microphones would work as well. [0] https://github.com/mozilla/DeepSpeech [1] https://wiki.seeedstudio.com/ReSpeaker_4_Mic_Array_for_Raspb...

How's the quality of Vosk compared to Mozilla DeepSpeech [1]? Most of offline the Speech to Text tools I know have a bad model that will have a hard time at understanding anything that is not English :( Mozilla DeepSpeech seems to have well trained models, from what I remember. [1]: https://github.com/mozilla/DeepSpeech

Mozilla announced Deep Speech[1] around the same time as Common Voice. Mozilla Deep Speech is an open source speech recognition engine, based upon Baidu's Deep Speech research paper[2]. Unsurprisingly, Deep Speech requires a corpus such as... Common Voice. [1] https://github.com/mozilla/DeepSpeech [2] https://arxiv.org/abs/1412.5567

If anyone is interested, Mozilla’s Deepspeech (https://github.com/mozilla/DeepSpeech) is also quite a solid open source offline alternative, and runs quite well even on low-end hardware like a Raspberry Pi.