Bulk Transcribe Audio to Text Online Free

Audio formats: MP3, WAV, M4A, FLAC, OGG, OPUS, AAC

Video formats: MP4, WebM, MKV, MOV, AVI (we extract the audio track)

Choose the format that fits your use case:

• TXT — Plain text with intelligent paragraph breaks (ML-powered)
• SRT — Subtitles with timestamps for video players
• VTT — Web captions for HTML5 video
• PDF — Professional document with AI title for sharing/archiving
• Word — Editable document with AI title
• Markdown — Clean format for note-taking apps

Automatic language detection via OpenAI Whisper. Both Fast and Quality models support 99+ languages including:

English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Japanese, Korean, Chinese, and many more.

Mixed-language content is handled automatically.

We offer two OpenAI Whisper models — Fast (Whisper V3 Large Turbo) for speed, and Quality (Whisper V3 Large) for maximum accuracy.

On clear recordings, it achieves near-human accuracy. Our ML post-processing structures the transcript into clean, readable paragraphs.

Results vary with audio quality — see our FAQ on accuracy limitations below.

No — and no transcription service can guarantee 100% accuracy.

Whisper achieves near-human accuracy on clear recordings in major languages. However, accuracy depends on:

• Audio quality and background noise
• Speaker clarity and accents
• Multiple overlapping speakers
• Domain-specific jargon and proper nouns

We recommend reviewing transcripts for critical use cases (legal, medical, published content).

After transcription, we run your text through a machine learning model trained specifically for paragraph segmentation.

It analyzes sentence structure and topic flow to determine natural paragraph breaks — turning a wall of continuous text into clean, readable paragraphs.

This works automatically for text and document outputs (TXT, PDF, Word, Markdown, EPUB); SRT and VTT formats retain timestamped segments for subtitle synchronization.

Duration limits depend on your plan:

• Free: 60 minutes (1 hours)
• Pro: 300 minutes (5 hours)

This accommodates everything from short clips to full podcasts and lectures.

Transcription costs 2 credits per minute of audio.

A 30-minute podcast costs 60 credits.

• Free: 100 credits daily (~50 minutes)
• Pro: 12000+ credits monthly (~100 hours)

• Plain Text — Raw transcript for scripts or simple reading
• SRT/VTT — Subtitles for video (SRT for editors, VTT for web)
• PDF — Fixed document to share or archive
• Word — Editable for corrections and collaboration
• Markdown — Notes apps (Obsidian, Notion)
• EPUB — E-readers (Kobo, Apple Books)

Two OpenAI Whisper models — choose speed or accuracy:

Fast is the default — great for clear audio, podcasts, and lectures.

Quality uses the full 1.55B-parameter model. Independent benchmarks show ~10% WER for Quality vs ~12% for Fast (Artificial Analysis). Choose Quality for accented speech, noisy recordings, or technical content.

Both models support 99+ languages. Switch in the options panel above.

Sources: Groq docs, Artificial Analysis benchmark, Hugging Face model cards.

Meeting Intelligence (speaker diarization) separates and labels different speakers in your audio.

When enabled, the transcript identifies who said what using labels like Speaker 1, Speaker 2, etc. Our AI post-processing then analyzes the context to infer actual names when possible — for example, if someone says "Hi, I'm John" or is addressed as "Thanks, Sarah."

Important: Name detection is not perfect. It works best when:

• Speakers introduce themselves at the start
• People are addressed by name during conversation
• Audio quality is clear with minimal background noise

Meeting Intelligence costs extra credits and has lower file size limits due to the additional processing required.

Speaker separation (diarization) is highly accurate when speakers have distinct voices and don't talk over each other. The AI correctly identifies different speakers in most clear recordings.

Name inference is more variable. The AI post-processing attempts to extract names from the audio context, but accuracy depends on:

• Whether speakers explicitly state their names
• How clearly names are pronounced
• Background noise and audio quality

If name detection misses someone, you'll see generic labels like "Speaker 1" — the transcript content is still accurate, just without the name attribution.