speechscore-metrics
Choose, validate, and run SpeechScore objective metrics for intrusive and non-intrusive speech quality assessment.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- speechscore-metrics
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Choose, validate, and run SpeechScore objective metrics for intrusive and non-intrusive speech quality assessment.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md5.4 KB
- 参考文档references/metric-catalog.md5.0 KB
- 参考文档references/troubleshooting.md5.4 KB
- 参考文档references/workflows.md5.1 KB
- 脚本scripts/speechscore_metric_recipe.py8.8 KB
03 SKILL.md 正文
SpeechScore Metrics
Use this sub-skill when a task asks for objective speech quality assessment with SpeechScore: selecting metrics, scoring one enhanced/degraded file against a clean reference, scoring matched test/reference directories, scoring reference-free MOS-style metrics, tuning window/score_rate/return_mean, or interpreting nested result dictionaries.
Start here:
- Read references/api-reference.md for the
SpeechScorefactory, import-layout rules, call parameters, result shapes, and importantwindow/score_ratecaveats. - Read references/metric-catalog.md before choosing metrics or deciding whether
reference_pathis required. - Read references/workflows.md for copyable dry-run, single-file, matched-directory, reference-free, and nested-result interpretation recipes.
- Read references/troubleshooting.md when imports, metric dependencies, model-backed metrics, references, directory basenames, or windowed scoring fail.
- Use scripts/speechscore_metric_recipe.py to validate requested metric names/reference behavior or to run scoring without assuming bundled sample paths.
Operating rules:
- Use this sub-skill only for SpeechScore objective metric selection and scoring.
- Route audio enhancement, separation, super-resolution, or target-speaker extraction generation to
../clearvoice-inference/before scoring. - Route benchmark list creation, dataset manifests, training launchers, or training/evaluation data preparation to
../training-and-data-prep/. - Decide
reference_pathfrom the metric catalog, not from each source class'sintrusiveattribute; that attribute is inconsistent in this snapshot. - For dry runs, validate metric names, whether any selected metric requires a reference, and matching directory basenames before reading audio.
- For real source-layout runs from outside the SpeechScore component directory, pass
--speechscore-dir <speechscore_component_dir>so the helper can importspeechscore.pyand resolve metric assets. - Prefer
window=Nonefor direct source API calls. Use the bundled helper when windowed scoring or explicit non-fixed metric resampling is required, because the inspected source window branch has amaxlenbug and the directscore_rateargument is not reliably honored for non-fixed metrics.
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).