Trait Synthesizer

Source

pub trait Synthesizer:
    Send
    + Sync
    + 'static {
    // Required methods
    fn sample_rate(&self) -> u32;
    fn synthesize_with(
        &self,
        text: &SpeechText,
        voice: Option<&Voice>,
        rate: SpeakingRate,
        out: SpeechWriter,
        cancel: &CancelToken,
    ) -> Result<(), SynthesisError>;

    // Provided method
    fn plays_audio_itself(&self) -> bool { ... }
}

Expand description

The contract every text-to-speech engine implements.

An implementation must:

write PCM into out as it is produced, so playback starts before the whole utterance is synthesized;
poll cancel at its natural pause points and answer SynthesisError::Cancelled;
treat a closed writer as the listener being gone and stop, answering SynthesisError::Cancelled;
close the writer on every exit — SpeechWriter::finish on success, SpeechWriter::fail with the cause on failure;
wrap its own failures in BackendError::engine under SynthesisError::Load or SynthesisError::Inference;
refuse a voice it does not have as SynthesisError::UnsupportedVoice.

Required Methods§

Source

fn sample_rate(&self) -> u32

The rate of the PCM this engine writes, fixed by its model.

Source

fn synthesize_with( &self, text: &SpeechText, voice: Option<&Voice>, rate: SpeakingRate, out: SpeechWriter, cancel: &CancelToken, ) -> Result<(), SynthesisError>

Synthesizes speech into out, chunk by chunk. voice of None means the engine’s configured default.

Provided Methods§

Source

fn plays_audio_itself(&self) -> bool

Whether this engine delivers sound itself rather than writing PCM: a platform voice program synthesizes and plays in one stroke. Synthesis::speak skips its own playback for such an engine. A wrapper around another synthesizer must forward this, or start hands back an empty session instead of refusing.