Output

File layout

source-project/
├── index.md                  ← injected <audio> tag
├── pipeline/
│   └── 01-parser.md          ← injected <audio> tag
├── index.mp3                 ← online TTS (edge-tts)
├── pipeline/
│   └── 01-parser.mp3         ← online TTS (edge-tts)
└── voice-notes/
    ├── index.md              ← voice-friendly rewrite
    └── pipeline/
        └── 01-parser.md      ← voice-friendly rewrite

For local TTS (Piper), .wav files replace .mp3.

Audio embed

Each processed .md file gets an <audio> tag injected at the top:

**Audio version** — Estimated duration: 2 min 59 sec
 
<audio controls src="https://storage.googleapis.com/project-breakdown101.firebasestorage.app/content/md2voice/reference/output.mp3">
  Your browser does not support the audio element.
</audio>
 

The embed skips files that already contain the **Audio version** marker. Duration is parsed via mutagen.

Voice notes

The LLM’s conversational rewrite is saved to voice-notes/ as a mirrored directory tree. These files contain:

• No markdown formatting
• No bullet points or tables
• Complete, natural sentences
• Verbal signposts (“First…”, “Let me explain…“)

Audio formats

Backend	Format	Bitrate/Sample rate
edge-tts	.mp3	Variable (Microsoft default)
Piper	.wav	16-bit mono PCM (model-dependent rate)