Voice MD

Voice-to-text transcription for Obsidian powered by OpenAI's GPT-4o audio models. Record audio, transcribe with automatic speaker identification, and optionally structure output with GPT formatting—all inline.

---

Quick Start

1. Install via Obsidian Community Plugins: Search "Voice MD"
2. Configure OpenAI API key in Settings → Voice MD (Get API key)
3. Record via ribbon icon or command palette (Start Voice Recording)

Text appears at cursor position. Optional meeting mode for speaker identification, post-processing for structured markdown.

---

Features

Core

• OpenAI transcription - gpt-4o-mini-transcribe (standard) / gpt-4o-transcribe-diarize (meeting mode)
• Inline insertion - Text inserted at cursor position in active note
• Auto-language detection - Or force specific language codes (en, es, fr, etc.)
• Mobile support - iOS and Android compatible (WebM/MP4/OGG/WAV)

Meeting Mode

Automatic speaker identification for multi-speaker recordings:

**Speaker A:** Welcome to today's meeting.
**Speaker B:** Thanks for having me. I've completed phase one.
**Speaker A:** Great. What were the main challenges?

• Toggle per-recording via checkbox in modal
• Uses gpt-4o-transcribe-diarize model
• Best with 2-6 speakers, recordings >30 seconds
• 10-30% additional response time

Post-Processing (Optional)

GPT-powered markdown formatting creates dual files:

• transcription-YYYY-MM-DD-HHmmss-raw.md - Original transcription
• transcription-YYYY-MM-DD-HHmmss.md - Formatted with heading hierarchy, lists, paragraphs

Features:

• Configurable model (gpt-4o-mini, gpt-4o)
• Custom system prompts for domain-specific formatting
• Preserves speaker labels from meeting mode
• Cross-linked files in Voice Transcriptions/ folder

---

Configuration

Settings → Voice MD

Setting	Description	Default
OpenAI API Key	Required. Get from platform.openai.com/api-keys	-
Max Recording Duration	Maximum seconds per recording	300
Auto-Start Recording	Start recording immediately on modal open	false
Language	Force language code (en, es, fr) or blank for auto-detect	Auto
Enable Post-Processing	GPT-powered markdown formatting	false
Chat Model	GPT model for post-processing	gpt-4o-mini
Custom Formatting Prompt	Override default formatting instructions	-

Note: Meeting Mode is enabled per-recording via checkbox in modal, not global settings.

---

Usage

Standard Workflow

1. Position cursor in note
2. Click microphone icon or run Start Voice Recording command
3. Speak naturally
4. Click stop
5. Text inserted at cursor

Meeting Mode

1. Open recording modal
2. Enable "Meeting Mode" checkbox
3. Record conversation
4. Speaker labels automatically added (**Speaker A:**, **Speaker B:**, etc.)

Post-Processing Output

When enabled, creates two files in Voice Transcriptions/:

• Raw transcription (timestamped, -raw.md suffix)
• Formatted version with structure (headings, lists, paragraphs)
• Formatted file includes link to raw: > Raw transcription: [[...]]
• Formatted text inserted at cursor

---

Best Practices

Audio Quality

• Speak clearly at moderate pace
• Minimize background noise for multi-speaker recordings
• Pause between major points for better segmentation

Meeting Mode

• Optimal: 2-6 speakers
• Minimum: 30 seconds for accurate identification
• Clear distinct voices improve accuracy

Post-Processing

• Verbally mention structure ("first point", "next section")
• Use custom prompts for specialized domains
• Review both raw and formatted versions

---

Installation

Community Plugin (Recommended)

1. Open Obsidian Settings → Community plugins
2. Browse and search "Voice MD"
3. Install and enable

Manual Installation

1. Download latest release from GitHub Releases
2. Extract main.js, manifest.json, styles.css to <vault>/.obsidian/plugins/voice-md/
3. Restart Obsidian
4. Enable in Settings → Community plugins

Beta Testing (BRAT)

For pre-release versions:

1. Install BRAT plugin
2. BRAT settings → Add Beta plugin → DenizOkcu/voice-md

---

Pricing

Requires OpenAI API key (pay-per-use):

• Transcription: gpt-4o-mini-transcribe / gpt-4o-transcribe-diarize
• Post-Processing: Configurable GPT model (gpt-4o-mini, gpt-4o)

View OpenAI API pricing →

---

Privacy & Security

Data Flow:

• Audio → OpenAI API (transcription only, ephemeral)
• Text → OpenAI API (post-processing only, if enabled)
• API key → Local Obsidian storage
• Recordings → In-memory processing (never written to disk)

No telemetry, tracking, or third-party services. See OpenAI Privacy Policy for API data handling.

---

Troubleshooting

Issue	Solution
"No API key configured"	Add OpenAI API key in Settings → Voice MD (Get key)
Recording doesn't start	Check microphone permissions: • macOS: System Settings → Privacy → Microphone • Windows: Settings → Privacy → Microphone
Transcription fails	• Verify API key validity • Check OpenAI account credits • Confirm internet connection • Check console logs (Cmd/Ctrl+Shift+I)
Meeting mode no speakers	• Recording >30 seconds • Improve audio quality • Verify meeting mode checkbox enabled
Post-processing fails	• Confirm enabled in settings • Verify API key model access • Check OpenAI credits • Review console errors

---

Contributing

Issues • Discussions

MIT License - see LICENSE