speaker_helper.cli module
Command-line interface for speaker-helper.
Module summary
One command line, built with click:
speaker-helper. It exposes the library as a terminal tool whose
sub-commands cover the three jobs the package exists for and the operations
around them:
synth— speak text (or stdin) into a.wav, offline or streaming, printing the measured duration and real-time factor.narrate— read a document aloud and write the WebVTT cues beside the audio, so a page can follow the reading.voices— list an engine’s preset voices.clone— register a cloned voice from reference audio (default: the bundled ref-fr-female) and print its id; a missing transcript comes fromvocal-helper.speak-from— re-voice audio from a YouTube URL, a podcast feed or the microphone; seespeaker_helper.sources.route— choose engine + mode from measured quality↔speed evidence.eval— measure the engine against a dataset and gate on versioned thresholds (exit0pass /1fail); seespeaker_helper.eval.install-voice— fetch an engine’s weights before anything depends on them.health— report whether the configured backend can run, and exit non-zero when it cannot.serve— run the REST API server; seespeaker_helper.api.
Global options (--backend, --engine, --language, the --clone
family, …) sit on the top-level group, so they precede the sub-command:
speaker-helper --backend mock synth "hi", speaker-helper --engine kokoro
voices. They are stashed on the click context and merged into the flag bag
each command handler reads.
Each _cmd_* handler takes that bag and returns a process exit code, so the
command logic stays testable without a terminal; the click layer above it only
parses.
Usage example
speaker-helper synth "Bonjour le monde." -o hello.wav --engine kokoro
echo "Texte depuis stdin." | speaker-helper synth -o out.wav
speaker-helper --engine kokoro voices
speaker-helper route --condition online_realtime --language fr
speaker-helper serve --bind-host 0.0.0.0 --bind-port 8080