microsoft/VibeVoice · error · RuntimeError

Voice preset {key!r} not found

Error message

Voice preset {key!r} not found

What it means

Raised by _ensure_voice_cached when asked to cache a voice preset key that is not in self.voice_presets. Preset keys are the .pt filename stems discovered at startup (e.g. 'en-Carter_man'), so any key from another naming scheme (path, display name, uppercase) will miss. Note the main request path (_get_voice_resources) already falls back to default_voice_key, so hitting this usually means an internal call passed a raw/unvalidated key.

Source

Thrown at demo/web/app.py:157

        print(f"[startup] Found {len(presets)} voice presets")
        return dict(sorted(presets.items()))

    def _determine_voice_key(self, name: Optional[str]) -> str:
        if name and name in self.voice_presets:
            return name

        default_key = "en-Carter_man"
        if default_key in self.voice_presets:
            return default_key

        first_key = next(iter(self.voice_presets))
        print(f"[startup] Using fallback voice preset: {first_key}")
        return first_key

    def _ensure_voice_cached(self, key: str) -> Tuple[object, Path, str]:
        if key not in self.voice_presets:
            raise RuntimeError(f"Voice preset {key!r} not found")

        if key not in self._voice_cache:
            preset_path = self.voice_presets[key]
            print(f"[startup] Loading voice preset {key} from {preset_path}")
            print(f"[startup] Loading prefilled prompt from {preset_path}")
            with torch.serialization.safe_globals([BaseModelOutputWithPast, DynamicCache]):
                prefilled_outputs = torch.load(
                    preset_path,
                    map_location=self._torch_device,
                    weights_only=True,
                )
            self._voice_cache[key] = prefilled_outputs

        return self._voice_cache[key]

    def _get_voice_resources(self, requested_key: Optional[str]) -> Tuple[str, object, Path, str]:
        key = requested_key if requested_key and requested_key in self.voice_presets else self.default_voice_key
        if key is None:

View on GitHub (pinned to 94da20d98b)

Solutions

1. Use the exact preset stem (filename without .pt), e.g. 'en-Carter_man', matching a file in demo/voices/streaming_model.
2. Print sorted(service.voice_presets) and pick the key from that list.
3. Set VOICE_PRESET to a valid stem or unset it to fall back to en-Carter_man / first key.
4. If presets changed on disk, restart the service so presets are re-discovered.

Example fix

# before
service._ensure_voice_cached("en-Carter_man.pt")

# after
key = "en-Carter_man"  # exact stem from service.voice_presets
service._ensure_voice_cached(key)

def valid_voice_key(service, key: str) -> bool:
    return key in service.voice_presets  # keys are .pt stems

if not valid_voice_key(service, requested):
    requested = service.default_voice_key

def is_known_voice(service, key: object) -> bool:
    return isinstance(key, str) and key in service.voice_presets

try:
    out = service._ensure_voice_cached(key)
except RuntimeError:
    key = next(iter(service.voice_presets))
    out = service._ensure_voice_cached(key)

When it happens

Trigger: Calling service._ensure_voice_cached('en-Carter_man.pt') (with extension), 'Carter', or a VOICE_PRESET env value that was never validated against the discovered keys; calling load-dependent methods before load() populated voice_presets.

Common situations: VOICE_PRESET env var set to a name that does not match any .pt stem in demo/voices/streaming_model; preset files renamed after startup; passing a Path instead of the stem string.

Related errors

• Voices directory not found: {voices_dir}
• No voice preset (.pt) files found in {voices_dir}
• Multiple voice presets match the speaker name '{speaker_name
• StreamingTTSService not initialized
• MODEL_PATH not set in environment

AI-assisted analysis of microsoft/VibeVoice@94da20d98b (2026-08-15). Data as JSON: /api/errors/85550e4078588a41. Report an issue: GitHub.