4 ms·
Author here. I think it's more of a capability issue than a safety issue. Since learning audio is still harder than learning text, audio models don't generalize
by vvolhejn 1y ago
Author here. I think it's more of a capability issue than a safety issue. Since learning audio is still harder than learning text, audio models don't generalize as well. To fix that, audio models rely on combining information from text and audio (having a single model that consumes/produces both text and audio tokens) and the audio tokens basically end up being an integrated speech-to-text/text-to-speech. This reflects my colleagues' experience working on Moshi, and it seems to be the case for other models too, see the Conclusion section.
Part of the reason can also be synthetic data: if you fine-tune on data generated from text via a text-to-speech, the tone of the voice doesn't have any information, so the model learns to ignore it.
- j45 1y agoAccent detection or consciously ignoring it is a filter step.
- JoshTriplett 1y agoAudio models for speech not understanding pitch, seems similar to how text LLMs often don't understand spelling: it's not what they were trying to recognize.
- oezi 1y ago> generated from text via a text-to-speech Yes, frustratingly we don't have good speech-to-text (STT/ASR) to transcribe such differences. I recently finetuned a TTS* to be able to emit laughter and hunting for transcriptions which include non-verbal sounds was the hardest part of it. Whisper and other popular transcription systems will ignore sigh, sniff, laugh, etc and can't detect mispronounciations etc. * = https://github.com/coezbek/PlayDiffusion https://github.com/coezbek/PlayDiffusion
- jasonjayr 1y agoIIRC -- the 15.ai dev was training on fan-made "My Little Pony" transcriptions, specificaly because they included more emotive clues in the transcription, and supported a syntax to control the emotive aspect of the speech.
- dotancohen 1y agoWhere can I read about this?
- jasonjayr 1y ago> During this phase, 15 discovered the Pony Preservation Project, a collaborative project started by /mlp/, the My Little Pony board on 4chan.[47] Contributors of the project had manually trimmed, denoised, transcribed, and emotion-tagged thousands of voice lines from My Little Pony: Friendship Is Magic and had compiled them into a dataset that provided ideal training material for 15.ai.[48] From https://en.wikipedia.org/wiki/15.ai#2016%E2%80%932020:_Conception_and_development https://en.wikipedia.org/wiki/15.ai#2016%E2%80%932020:_Conce...
- vvolhejn 1y agoI had no idea this existed, the internet is amazing
- smusamashah 1y agoThere was an example, of ChatGPT copying and responding in the speakers voice mid conversation, on OpenAI blog. This was presented an example on non-alignment.
- wordglyph 1y agoI used aistudio and it understood pitch and and even emotion with an uploaded mp3