How to transcribe audio
- Tick the consent box, then choose a file (MP3, M4A, WAV, video…) or press Record now.
- Your browser shrinks the audio to 16 kHz mono (speech models do not need more), then sends it for transcription.
- Copy the text, download a .txt or .srt, or press Write it for show notes, a summary with action items, a cleaned-up version, or title ideas.
Getting the most accurate transcript
- Record close to the mic in a quiet room. Background music is the biggest accuracy killer.
- Clean noisy files with the noise remover before transcribing.
- For interviews, give each person their own mic if you can; overlapping speech is hard for any model.
- Split recordings longer than 10 minutes into parts with the trimmer.
Consent and privacy
Only transcribe recordings you made or have permission to use. Recording laws differ: many US states and several countries require every person in a conversation to agree to being recorded. VocalForge does not identify speakers and does not keep your audio.
Questions
Is the transcription free?
Yes, up to 8 transcriptions of 10 minutes each per hour. There is no account.
Which languages are supported?
The speech model recognises dozens of languages and detects the language automatically, including English, Spanish, French, German, Portuguese, Hindi, Japanese and Chinese. Accuracy is highest for clear speech close to the mic.
Is my audio stored?
VocalForge does not store it. The compressed audio is sent once through Vercel AI Gateway to the speech model and the text comes back to your browser. Your last 10 transcripts are saved only in this browser's local storage, and you can clear them at any time.
Can I get subtitles?
Yes. When the model returns timestamps, you can download an .srt subtitle file that YouTube, Premiere, DaVinci Resolve and VLC accept.
How accurate is it?
For a clean single-speaker recording, expect very few errors. Crosstalk, heavy accents, music beds and distant mics all reduce accuracy. Running the file through the noise remover first can help.