Guard against exceeding the maximum TextDecoder token sequence length for non-streaming mode

argmaxinc / WhisperKit

On-device Speech Recognition for Apple Silicon

https://takeargmax.com/blog/whisperkit

MIT License

3.17k stars 268 forks source link

Guard against exceeding the maximum TextDecoder token sequence length for non-streaming mode #89

Closed atiorh closed 5 months ago

atiorh commented 6 months ago

Guard against exceeding the maximum TextDecoder token sequence length for non-streaming mode. Long and fast enough speech within a 30-second window occasionally requires more tokens than maximum allowed in Whisper (448 total count, 445 excluding special task tokens)

ZachNagengast commented 5 months ago

Closing based on this issue https://github.com/argmaxinc/WhisperKit/issues/63, can reopen either if it shows up again