voxr 0.7.0

Changed

Dependencies

• voxr and voxr-narrate require boostr 0.9 and numr 0.13. voxr's public fields name boostr model types, such as WhisperBundle::model, and its engines take numr runtime types.
• voxr-narrate requires voxr 0.7.

Full diff: v0.6.0...v0.7.0

voxr 0.6.0

Changed

Dependencies

• voxr and voxr-narrate require boostr 0.8. voxr's public fields name boostr model types, such as WhisperBundle::model, and voxr-narrate's adapters name boostr types.
• voxr requires splintr 0.22. WhisperBundle::tokenizer is a splintr AnyTokenizer.
• voxr-narrate requires voxr 0.6.

Full diff: v0.5.0...v0.6.0

voxr 0.5.0

Changed

Dependencies

• voxr and voxr-narrate require boostr 0.7 and numr 0.12. voxr's engines and scorers take numr runtime types, and voxr-narrate's adapters name boostr and numr types. A caller that hands voxr a tensor or runtime uses these versions.
• voxr-narrate requires voxr 0.5.

Full diff: v0.4.0...v0.5.0

voxr 0.4.0

Added

voxr

• KokoroEngine::from_dir_on and VitsEngine::from_dir_on load an engine onto any numr runtime's device, such as CUDA.
• load_vits_from_dir_on builds the VITS TtsBundle on any runtime's device.
• KokoroEngine::synthesize_seeded and KokoroEngine::with_seed seed Kokoro's harmonic source. The same text, voice, speed and seed give the same samples on one device. CPU and GPU run separate PRNGs, so one seed gives different samples on each.

voxr-narrate

• KokoroSpeech passes the take's seed to Kokoro. A re-roll at a new seed now gives a new Kokoro take.

Changed

voxr

• KokoroEngine<R> and VitsEngine<R> are generic over the runtime, with CpuRuntime as the default. from_dir still loads on the CPU, so existing call sites compile unchanged. KokoroEngine's model, client and device fields take R's types.
• KokoroEngine holds a private seed field, so a struct literal no longer builds one. Use from_dir or from_dir_on.
• TtsEngine holds for both engines on every runtime whose client covers the model's ops.
• A seeded VITS take reproduces on one device only. CPU and GPU draw different noise from one seed.

voxr-narrate

• KokoroSpeech<'a, R> and VitsSpeech<'a, R> are generic over the runtime, with CpuRuntime as the default.
• The kokoro and vits features enable the numr and boostr dependencies, whose types the adapters' bounds name.
• KokoroSpeech refuses a seed that is not a whole number from 0 to 2^53 - 1, as VitsSpeech does.

Dependencies

• voxr and voxr-narrate require boostr 0.6 and numr 0.11. boostr 0.6 runs Kokoro on any runtime, and numr 0.11 makes the inverse STFT give the same samples on every GPU run. The Tensor and runtime types voxr passes through come from these versions, so a caller that hands voxr a tensor must use the same ones.

Full diff: v0.3.0...v0.4.0

voxr 0.3.0

Added

voxr-narrate

• AnchorStep::ShortOpening: a voice designed from an opening of fewer than OPENING_WORDS words says that its reference is short.
• Doubtful and doubtful_parts: the parts a speech check marks ok ?, counted by kind (doubtful word, stray pause, pitch peak, click).

Changed

voxr-narrate

• verdict_lines takes the Doubtful count beside the problems. The summary names the parts marked ok ?, with their part numbers and kinds. The exit code stays 0.
• AnchorStep has a new variant, so a match over it needs an arm for ShortOpening.
• A VoxCPM2 stray pause advises --reroll for its part. Each stray pause adds BAD_STRAY_PAUSE to a take's badness, so retries and re-rolls keep the take with fewer.

Full diff: v0.2.0...v0.3.0

voxr 0.2.0

Changed

• Dependencies — voxr and voxr-narrate require boostr 0.5 and numr 0.10. The Tensor and runtime types they pass through come from these versions, so a caller that hands voxr a tensor must use the same ones.

Full diff: v0.1.1...v0.2.0

voxr 0.1.1

Fixed

voxr-narrate

• Compiling the narration patterns no longer overflows a 1 MiB stack, the size of a Windows main thread. A debug build crashed on the first script it read.

Full diff: v0.1.0...v0.1.1

voxr 0.1.0

First release of voxr and voxr-narrate.

Added

voxr

• RIFF/WAVE read and write, including stream headers of unknown length.
• Compressed-audio decode behind decode: FLAC, MP3, AAC-LC and ALAC in MP4/M4A/MOV, and Opus and Vorbis in Ogg/WebM/MKV, with ffmpeg-matched trims and downmix. Multichannel AAC, HE-AAC and Opus past two channels are refused.
• Resamplers: a general polyphase resampler, and resamplers that match ffmpeg's default output, SciPy's resample_poly and librosa's default output.
• A WSOLA tempo engine that matches ffmpeg's atempo output.
• The reference-take enhancement chain: denoise, filters, loudness normalization and a limiter.
• Pitch tracking, take quality, spectral distance, and word and character error rates.
• Single-sound metrics and the sound review of the motion-video skill, with CLAP ranking behind clap.
• English G2P: the misaki lexicon pipeline behind misaki, the spaCy en_core_web_sm tokenizer behind spacy-tokenizer, its tagger behind spacy-tagger, and a neural out-of-vocabulary fallback.
• TtsEngine, TtsBundle and Voice behind tts.
• Kokoro synthesis behind kokoro, loaded from a model directory.
• VITS (MMS-TTS) synthesis behind vits, with uroman romanization for is_uroman checkpoints.
• VoxCPM2 voice cloning behind voxcpm: reusable reference voices, continuation prompts, per-request guidance and steps, bad-take retry and batched requests.
• Whisper transcription with word timings behind whisper.
• Silero VAD segmentation behind vad.
• Speech-corpus preparation behind corpus.
• The TTS quality gate: signal statistics, pace, intelligibility and timbre.

voxr-narrate

• Narration script, timed script and subtitle parsing, tags, and numbers read back from words and digits.
• The speech check of a take against its script, and glitch detection and repair.
• The voice polish chain, and reference-voice preparation for cloning.
• The take loop: cache keys, seeds, take records, the pick of each part's best take, speaking pace and its verdict, anchor voices, timed placement and joins.
• A whole VoxCPM2 run from script to voiceover, and a per-phrase run for every other engine.
• Engine adapters for Kokoro, VITS and VoxCPM2, and Whisper as the hearing, behind features.
• Time-stretch of the spoken track to a pace behind stretch.
• The speak and transcribe commands of the motion-video skill: argument parsing, every printed line, file and exit code, and the video manifest's narration.

Tree: https://github.com/ml-rust/voxr/tree/v0.1.0