pocketsphinx VS AudioWorkletStream

There are espeak-ng https://github.com/espeak-ng/espeak-ng and pocketsphinx https://github.com/cmusphinx/pocketsphinx which can be used locally without making external requests.

And so I installed PocketSphinx, "one of Carnegie Mellon University's open source large vocabulary, speaker-independent continuous speech recognition engines."

- https://github.com/cmusphinx/pocketsphinx

This avoids having to stream audio 24x7 to a cloud model which would be super expensive. This being said, I'm pretty sure what the Alexa does, for example, is send any positive wake word to a cloud model (that is bigger and more accurate) to verify the prediction of the local wake word detection model AFAIK.

PocketSphinx is generally regarded among voice assistant communities as a less reliable, but straight OOTB, alternative to a robust listener. It's a good solution when you want multiple hotwords (or just aren't in a position to train even one word.)

// Exits half way through reading response when --max-old-space-size=6 is set // Exits immediately when --jitless flag is set // // Usage: // // port.postMessage({ // url: 'https://github.com/guest271314/AudioWorkletStream/raw/master/house--64kbs-0-wav', // method: 'get', // body: null // })

Yes. Fetching a single or multiple files for an infinite stream of audio https://github.com/guest271314/AudioWorkletStream/blob/master/worker.js. Streaming (real-time) audio is non-trivial. Any gaps or glitches in playback will be audible to the user. We could test for expected Float32Arrays. I would suggest complentary manual test in, e.g., WPT to determine audio output does not have gaps or glitches; and renders the expected playback rate.

One way to do that is utilizing Native Messaging on Chromium or Firefox https://github.com/guest271314/native-messaging-espeak-ng, and https://github.com/guest271314/webtransport/blob/main/webTransportEspeakNg.js for some WebTransport experiments. Technically we don't need a local server. We can stream and parse the WAV directly and pipe to AudioWorklet or a MediaStreamTrackGenerator https://github.com/guest271314/AudioWorkletStream. The same is true for speech recognition, where audio is piped to the local application and text or JSON piped back. Note also espeak-ng has been compiled to WebAssembly. I created native-messaging-espeak-ng for the ability to pass SSML directly to espeak-ng.

You can stream audio and/or video over HTTP using fetch() https://github.com/guest271314/AudioWorkletStream as long as you know how to parse the codec, if the media is encoded.

Re using a dedicated Worker and AudioWorklet to stream, see, e.g., https://github.com/guest271314/AudioWorkletStream; https://plnkr.co/edit/nECtUZ.