org.llm4s.speech

Output representation for generated audio.

Audio input representations used by speech components.

Basic audio metadata.

For AudioFormat.Mp3 audio bitDepth is 0, meaning "not applicable": MP3 is compressed, so it has no sample width, and a duration or frame count cannot be computed from data.length and these fields. sampleRate and numChannels are then the service's nominal values for the format it was asked for, not read from the stream.

Routes speech provider configuration to the appropriate TTS or STT client.

Provider selection follows the same provider/model prefix pattern used for LLM providers. The model format, set through SPEECH_TTS_MODEL / SPEECH_STT_MODEL (see org.llm4s.speech.config.SpeechConfigLoader), is: openai/tts-1, elevenlabs/ , azure/ (TTS) openai/whisper-1, azure/en-US (STT)