Skip to content

Credential-Free Voice Demo

director_ai.voice.demo provides a deterministic dry-run voice pipeline for local validation, CI, and onboarding. It uses the production async voice_pipeline() plus DryRunTTSAdapter, so the guard, sentence buffering, halt callback, recovery path, and adapter lifecycle are exercised without external services or credentials.

from director_ai.voice import run_voice_demo

result = await run_voice_demo(use_nli=False)

print(result.tts_texts)
print(result.audio_chunks)
print(result.total_audio_bytes)

Run it from the command line:

python -m director_ai.voice.demo

The dry-run adapter records every text fragment sent to synthesis and emits deterministic bytes prefixed with audio:. This makes it suitable for regression tests and for validating that a deployment path closes adapters on all completion and failure paths before replacing the adapter with a real TTS backend.

director_ai.voice.demo.DryRunTTSAdapter dataclass 

DryRunTTSAdapter(prefix: bytes = b'audio:', texts: list[str] = list(), closed: bool = False)

Bases: TTSAdapter

TTS adapter that records text and yields deterministic audio bytes.

synthesise async 

synthesise(text: str) -> AsyncIterator[bytes]

Record text and emit deterministic dry-run audio bytes.

close async 

close() -> None

Mark the dry-run adapter as closed.

director_ai.voice.demo.VoiceDemoResult dataclass 

VoiceDemoResult(audio_chunks: list[bytes], tts_texts: list[str], halted: bool, halt_reason: str = '', recovery_text: str = '')

Summary returned by :func:run_voice_demo.

total_audio_bytes property 

total_audio_bytes: int

Total number of dry-run audio bytes emitted by the demo.

director_ai.voice.demo.run_voice_demo async 

run_voice_demo(*, tokens: AsyncIterator[str] | Iterable[str] | None = None, facts: dict[str, str] | None = None, prompt: str = DEFAULT_PROMPT, tts: DryRunTTSAdapter | None = None, threshold: float = 0.3, hard_limit: float = 0.25, score_every: int = 4, soft_halt: bool = True, recovery: str = 'Let me verify that before speaking.', use_nli: bool = False) -> VoiceDemoResult

Run the package voice demo against the production async pipeline.

Parameters mirror :func:voice_pipeline; defaults are credential-free and deterministic so the demo is suitable for CI and quick local validation.

director_ai.voice.demo.scripted_tokens async 

scripted_tokens(text: str, *, separator: str = ' ') -> AsyncIterator[str]

Yield deterministic token-like fragments from text.

The default keeps whitespace attached to each token so sentence buffering sees realistic fragments while remaining stable across Python versions.