voice-goio docs

voicecheck CLI

cmd/voicecheck is the library’s regression and audit tool. Run it from the repository root:

go run ./cmd/voicecheck <command> [flags]
go run ./cmd/voicecheck <command> -h   # the flags of one command

Every command that reads voices takes -voices (a manifest path; the embedded voices.json by default) and -dir (the voice model folder; voices.Dir() by default).

Command	What it does
`synth`	Synthesise the audit text in every voice and radio profile, one WAV each.
`audit`	Render the audit text in every speaker of one non-English model, for the swap-trick listening pass.
`recog`	Replay a recognition corpus and compare the tags.
`devices`	List the audio input and output devices.
`voices`	Report the pool size, accent coverage and missing models against the bar.
`download`	Fetch models from the voice repository.
`pack`	Zip a voice pack of the installed models, for an installer.

synth

go run ./cmd/voicecheck synth -tts piper -limit 0 -profiles tower,center

Flag	Default	Meaning
`-text`	`testdata/audit/clearance.txt`	The text to speak.
`-out`	`wav`	Output directory.
`-tts`		Backend: `piper`, `say` or `fake`.
`-limit`	`12`	Maximum number of voices (0 = no limit).
`-profiles`	all	Comma-separated radio profiles.

audit

go run ./cmd/voicecheck audit -model de_DE-thorsten-medium

-model is required. -text, -out (default wav-audit) and -tts as for synth. The audit fields in the manifest (quality, pass) are filled in by the human listening pass this renders for; see Voice assignment.

recog

go run ./cmd/voicecheck recog                  # the corpus on stt/fake
go run ./cmd/voicecheck recog -backend sapi -wav wav-recog

Flag	Default	Meaning
`-backend`	`fake`	`fake` or `sapi`.
`-script`	`testdata/stt/readbacks.jsonl`	The corpus.
`-min`	`95`	Minimum percentage of exactly tagged phrases; below it the command fails.
`-wav`		Render each phrase to a WAV here and feed the file to the engine (unattended SAPI runs).
`-tts`		The backend that renders those phrases.

voices

go run ./cmd/voicecheck voices

Fails below -min assignable voices (default 100, the BeyondATC “Basic” bar) or -min-accents distinct accents (default 10).

download

go run ./cmd/voicecheck download -model en_GB-vctk-medium   # one model
go run ./cmd/voicecheck download -pack en                    # a pack
go run ./cmd/voicecheck download -model all -write voices/voices.json

One of -model (a name, or all) or -pack (core, en, all) is required. -base points at another base URL or a LAN mirror; -write writes the manifest back with the SHA-256 hashes filled in.

pack

go run ./cmd/voicecheck pack -pack en -out voices-en.zip

Writes the zip an installer unpacks into the voices folder: each model’s .onnx and config at their manifest paths, plus a voices.json of just those models. Every model in the pack must be installed in -dir; a pack is whole or not made. -pack is core, en or all (the default); -out is required. In code this is Manifest.Pack and Manifest.WritePack.