buchwandler

Voices and discovery

TTSForge does not ship a fixed voice list. The available voices, references, models, and runtime status come from Readio’s discovery catalog and depend on the installed engine integrations and selected model/profile.

List voices

ttsforge voices
ttsforge voices --language en-us
ttsforge voices --engine kokoro
ttsforge voices --engine piper --model MODEL_ID

Filters include --engine, --language, --model, and --gender. Use --offline to avoid network discovery and --refresh to refresh cached catalog data. --json returns catalog items together with discovery information.

Human-readable output shows the canonical Readio voice ID (VoiceInfo.id) as the primary name, with the semantic reference (VoiceInfo.ref) when available. If no reference is available, the qualified voice ID can be used. also include language/locale, model, engine, status, and runtime availability. Readio owns these identities; do not assume legacy Kokoro IDs or language-prefix conventions apply to every engine.

Inspect models and engines

ttsforge models
ttsforge models --engine kokoro --language en-us
ttsforge engines
ttsforge doctor

models reports Readio model/synthesis targets and their status. engines lists registered synthesis engines and their installed/runnable state. doctor includes missing runtime dependencies and format availability.

Use a discovered voice

For explicit --voice values, use the canonical voice ID or semantic reference reported by Readio:

ttsforge convert novel.epub --engine kokoro --voice af_heart
ttsforge preview novel.epub --engine kokoro --voice af_heart

Guided selection during conversion

ttsforge convert uses the same Readio discovery services interactively on a TTY. It shows filtered choices as compact vertical lists. Enter a row number or exact identifier. Voice entries show the canonical Readio voice ID first and the semantic reference (VoiceInfo.ref) when available. Guided voice prompts accept a row number, canonical voice ID, semantic reference, or qualified ID, then keep the selected row’s canonical ID. Catalogs are filtered by language and engine, and by model for voices. Explicit --model and --voice options bypass matching prompts.

The engine-specific selection layouts are:

  • Kokoro: model, then voice.

  • Piper: choose a voice-bundle target. The matching voice has the same canonical identity and is selected automatically, without a duplicate prompt.

  • Pocket: choose a model bundle, then a predefined named voice.

The completed selection is strictly resolved by Readio after catalog choices are gathered. Pocket speed defaults to 1.0. TTSForge does not expose Readio’s Pocket reference WAV voice_file option in ttsforge convert. Voice and model details remain engine-specific. TTSForge does not implement voice blending or maintain voice recommendations. See Readio’s catalog documentation for the public discovery contract.

Selection layouts

The interactive converter follows each engine’s public Readio catalog semantics:

Kokoro: model -> voice
Piper:    voice-bundle target
Pocket:   bundle -> predefined named voice

For Piper, the selected bundle is stored in Readio’s neutral model request field, and its matching canonical voice is selected automatically. Pocket stores the bundle in model and the chosen predefined voice in voice.