Speech-to-Text (STT)

Speech-to-Text (STT) functionality for fmus-vox.

This module provides functionality for transcribing speech to text using various models and techniques.

fmus_vox.stt.transcribe(audio: str | Audio, model: str = 'whisper', language: str | None = None, **kwargs) → str

Transcribe audio to text using a specified model.

This is a simple functional API for quick transcriptions. For more control, use the Transcriber class directly.

Parameters:

• audio – Audio to transcribe (file path or Audio object)

• model – Model to use for transcription (whisper, wav2vec, etc.)

• language – Language code (if None, auto-detect)

• **kwargs – Additional model-specific parameters

Returns:

Transcribed text

Raises:

TranscriptionError – If transcription fails

Examples

>>> # Transcribe an audio file
>>> text = transcribe("recording.wav")
>>> print(text)

>>> # Transcribe with a specific model and language
>>> text = transcribe("recording.wav", model="whisper-large", language="en")

Transcriber Class

class fmus_vox.stt.transcriber.Transcriber(model: str = 'whisper', **kwargs)

Bases: object

Base class for speech-to-text transcription.

This class provides the common interface for all transcription models and handles model loading, caching, and transcription.

Parameters:

• model – Name of the model to use (whisper, wav2vec, etc.)

• device – Computation device (cpu, cuda, auto)

• **kwargs – Additional model-specific parameters

classmethod register_model(name: str, implementation: type) → None

Register a model implementation.

Parameters:

• name – Model name

• implementation – Model implementation class

static __new__(cls, model: str = 'whisper', **kwargs) → Transcriber

Create a new Transcriber instance of the appropriate subclass.

Parameters:

• model – Name of the model to use

• **kwargs – Additional model-specific parameters

Returns:

Transcriber instance

Raises:

ModelError – If the model is not supported

__init__(model: str = 'whisper', device: str | None = None, **kwargs)

Initialize the transcriber.

Parameters:

• model – Name of the model to use

• device – Computation device (cpu, cuda, auto)

• **kwargs – Additional model-specific parameters

transcribe(**kwargs)

transcribe_with_metadata(**kwargs)

async transcribe_async(audio: str | Audio, language: str | None = None) → str

Transcribe audio to text asynchronously.

Parameters:

• audio – Audio to transcribe (file path or Audio object)

• language – Language code (if None, auto-detect)

Returns:

Transcribed text

Raises:

TranscriptionError – If transcription fails

async transcribe_with_metadata_async(audio: str | Audio, language: str | None = None) → TranscriptionResult

Transcribe audio to text asynchronously with additional metadata.

Parameters:

• audio – Audio to transcribe (file path or Audio object)

• language – Language code (if None, auto-detect)

Returns:

TranscriptionResult object

Raises:

TranscriptionError – If transcription fails

stream(audio_stream: Generator[Audio, None, None], language: str | None = None) → Generator[TranscriptionResult, None, None]

Stream transcription for incoming audio chunks.

Parameters:

• audio_stream – Generator yielding Audio objects

• language – Language code (if None, auto-detect)

Yields:

TranscriptionResult for each processed chunk

Raises:

TranscriptionError – If transcription fails

Whisper Transcriber

Whisper model implementation for speech-to-text.

This module provides the WhisperTranscriber class which uses OpenAI’s Whisper model for transcription.

class fmus_vox.stt.whisper.WhisperTranscriber(model: str = 'whisper', **kwargs)

Bases: Transcriber

Transcriber using OpenAI’s Whisper model.

Whisper is a general-purpose speech recognition model that can transcribe speech in multiple languages and translate it to English.

Parameters:

• model – Whisper model size/variant (tiny, base, small, medium, large)

• device – Computation device (cpu, cuda, auto)

• download_root – Directory to download and store models

• **kwargs – Additional model-specific parameters

__init__(model: str = 'whisper-base', device: str | None = None, download_root: str | None = None, **kwargs)

Initialize the Whisper transcriber.

Parameters:

• model – Whisper model size/variant (tiny, base, small, medium, large)

• device – Computation device (cpu, cuda, auto)

• download_root – Directory to download and store models

• **kwargs – Additional model-specific parameters

transcribe_with_metadata(audio: str | Audio, language: str | None = None) → TranscriptionResult

Transcribe audio to text with additional metadata.

Parameters:

• audio – Audio to transcribe (file path or Audio object)

• language – Language code (if None, auto-detect)

Returns:

TranscriptionResult object

Raises:

TranscriptionError – If transcription fails

stream(audio_stream: Generator[Audio, None, None], language: str | None = None) → Generator[TranscriptionResult, None, None]

Stream transcription for incoming audio chunks.

Parameters:

• audio_stream – Generator yielding Audio objects

• language – Language code (if None, auto-detect)

Yields:

TranscriptionResult for each processed chunk

Raises:

TranscriptionError – If transcription fails