inference
Routes VITS text-to-speech synthesis and voice-conversion tasks from checkpoints.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- inference
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Routes VITS text-to-speech synthesis and voice-conversion tasks from checkpoints.
- disable-model-invocation
- true
02 资源文件 1
- 参考文档references/troubleshooting.md1.8 KB
03 SKILL.md 正文
Inference
Use this route when you need to synthesize speech from a checkpoint or run voice conversion with a VITS model.
Use this route when
- You have a checkpoint and want generated audio.
- You need to choose synthesis settings such as noise scale, length scale, or speaker id.
- You need the LJ Speech single-speaker inference path or the VCTK multi-speaker path.
- You need voice conversion from a source audio file or spectrogram to a target speaker.
- You need to troubleshoot checkpoint mismatches, sample-rate mismatches, or speaker-id problems.
Do not use this route when
- You are preparing filelists or building the extension; use
data-preparation. - You are launching training or resuming checkpoints; use
training. - You only need a quick environment sanity check; use
../../scripts/check_install.pyor../../scripts/model_smoke.py.
Read first
../../references/workflows.mdfor the synthesis and voice-conversion command flow.../../references/configuration.mdfor the config and file-layout differences.../../references/api-reference.mdforSynthesizerTrn.inferandvoice_conversion.../../references/troubleshooting.mdfor cross-cutting failures.references/troubleshooting.mdin this sub-skill for checkpoint and speaker-id failures.
Bundled helpers
../../scripts/synthesize.py— run TTS or voice conversion from a checkpoint.../../scripts/model_smoke.py— validate inference and voice-conversion wiring on synthetic inputs.../../scripts/check_install.py— confirm the repo imports and the backend is ready.
Common workflow
- Confirm the checkpoint matches the config family.
- Build the monotonic-alignment extension and verify CUDA.
- For TTS, provide text and speaker id when the config is multi-speaker.
- For voice conversion, provide source audio and both source and target speaker ids.
- Use
model_smoke.pywhen you need to check the model path without a real checkpoint.
Ownership boundaries
- Include checkpoint-driven TTS, voice conversion, and audio-output planning here.
- Exclude dataset cleanup and training launch; route those to sibling sub-skills.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).