Skip to content

Add pipeline benchmarking script - #14

Open
gino287 wants to merge 4 commits into
v0.4.1_backupfrom
codex/create-voice-pipeline-evaluation-script
Open

Add pipeline benchmarking script#14
gino287 wants to merge 4 commits into
v0.4.1_backupfrom
codex/create-voice-pipeline-evaluation-script

Conversation

@gino287

@gino287 gino287 commented Jul 30, 2025

Copy link
Copy Markdown
Contributor

Summary

  • create pipeline_benchmark.py example for running separation, speaker ID, and ASR in a batch
  • measure timing, SI‑SDR, CER, and accuracy metrics
  • output results to CSV and display overall summary

Testing

  • python -m py_compile examples/pipeline_benchmark.py

https://chatgpt.com/codex/tasks/task_e_688a4bf05164832a80ecef249334ea18

gino287 added 3 commits July 31, 2025 15:54
Replaces the old mixture_map.csv with a new format and location, and removes the pipeline_benchmark.py example script. The AudioSeparator class is simplified by removing the speaker count detection logic and related dependencies, streamlining the separation process. Also updates default Whisper model to 'large-v3' and beam size to 3 in constants.
manual_add_speaker.py:用於手動新增語者識別資料。

pipeline_eval.py:可批次執行語音處理流程,並計算各項評估指標。

predefined_speaker_builder.py:根據指定音檔自動建立語者聲紋檔。

此外,mixture_map.csv 現已更新為符合新的檔名命名規則。

重構了 pipelines/orchestrator.py,
改為支援顯式模組初始化與注入(module injection),
提升流程的彈性與 CLI 執行的可用性。
重構 pipeline_eval.py,改善 CER 計算與預測分配邏輯

新增 examples/stats_summarizer.py,用於彙整與總結 pipeline 評估指標

更新 WhisperASR 預設語言為中文 (zh)

優化 orchestrator.py 的模組初始化邏輯,增加向下相容性

調整語者辨識的閾值,並在 constants.py 更新 Whisper 的預設模型
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant