# Voice memo practice input

Original script authored for ZL-179 on 2026-09-10. No customer data and no third-party recording. Speech is synthetic (espeak-ng `en-us` at 145 wpm), then wrapped as AAC inside M4A. The reference is an authored dictation, not a transcription-provider output or an accuracy claim.

Reproduce on Linux with espeak-ng and FFmpeg:

```sh
espeak-ng -v en-us -s 145 -w /tmp/zl179-harbor-line.wav -f public/examples/voice-memo-to-text/reference.txt
ffmpeg -y -i /tmp/zl179-harbor-line.wav -c:a aac -b:a 64k -movflags +faststart harbor-line-memo.m4a
```

Committed M4A: 43.344 seconds, 408,134 bytes, AAC mono 22.05 kHz. Users may download and reuse the sample to test the tool. Playing the sample does not start a transcription request. Recognition, noise robustness, speaker detection, and live billing have not been benchmarked with this clip.