# VoxStudio 声音库自动识别人物测试 / Voice-library person recognition test

测试日期 / Tested: 2026-10-21. 测试者 / By: VoxStudio Team (the app developer).

## 测试版本与设备 / App build and machine

- VoxStudio 7.0.45 (131), a pre-release build. Re-check on the public release before publishing version-specific claims.
- Apple M5 MacBook Pro, 32 GB memory, macOS 27.0.1 (26A434).

## 设置 / Setup

1. Enrollment: in the English two-speaker session (`Two-speaker interview test · English (run 3)`), Speaker 1 (LibriSpeech speaker 1320) was saved from the transcript label menu with "Save voice to library" as a new person, `Test Person A`. The dialog reported 30.0 s of clear speech.
2. Test file: `en-heldout-three-voice.wav`, 118.7 s, three LibriSpeech voices and 15 utterances in 12 turns. Voice A is speaker 1320 with utterances that were not used for enrolment. Voices B (1995) and C (1089) were never enrolled. See `en-heldout-manifest.json`.
3. Processing options: Input language English, Speaker count Auto-detect, Recognize people = `Test Person A` only, Segment subtitles off.

## 结果 / Result

| Measure | Value |
| --- | --- |
| Utterances from the enrolled voice (A) | 5 |
| Utterances named `Test Person A` | 5 (all from A) |
| Utterances from non-enrolled voices (B, C) | 10 |
| False names (B or C named `Test Person A`) | 0 |
| Labels in the export | `Test Person A`, `Speaker 2`, `Speaker 3` (each label maps to one voice) |

Export: `heldout-run1-labelled.txt` (Transcript · Original · TXT · With speaker labels). Scored with `score_named.py`.

## 限制 / Limits

- One test file, one enrolled person and two non-enrolled voices. This is a worked example, not an accuracy claim.
- Read speech only. No overlap, interruption or background noise.
- Automatic naming in this build applies to local transcription (the matching code requires local compute). Manual links always win.
- The matching thresholds in this build are provisional and have not been calibrated across many people or recordings.
- Not tested: more than two non-enrolled voices, similar-sounding people, several enrolled people, cross-language enrolment, the Mandarin file, other devices.

## 数据位置 / Where the library lives

The people list is stored in the app's Application Support folder on the Mac (`VoxStudio/VoiceLibrary/persons.json`).