Scribe v1 Speech to Text

Scribe v1 is ElevenLabs' state-of-the-art automatic speech recognition (ASR) model. It achieves 96.7% accuracy in English and consistently outperforms leading models like OpenAI Whisper v3, Google Gemini 2.0 Flash, and Deepgram Nova-3 across 99 languages.

Scribe v1 achieves a word error rate (WER) of just 3.3% in English and 1.3% in Italian according to FLEURS benchmarks. This translates to approximately 96.7% accuracy, making it the most accurate publicly available ASR model.

Scribe v1 supports 99 languages with automatic language detection. It also handles code-switching, meaning it can accurately transcribe audio that switches between different languages within the same recording.

Scribe v1 can identify and label up to 32 different speakers in a single recording. Each speaker is labeled accurately, making it ideal for complex meetings, panel discussions, and multi-participant conversations.

Audio event tagging automatically detects and labels non-verbal sounds in your transcription, such as (laughter), (applause), (footsteps), or (music). This adds valuable context that pure speech transcription misses.

Scribe v1 supports all major audio and video formats including MP3, WAV, AAC, M4A, OGG, MP4, WebM, and more. The maximum file size is 3GB.

In benchmark tests (FLEURS & Common Voice), Scribe v1 consistently outperforms OpenAI Whisper Large v3 across all 99 supported languages, with particularly significant improvements in accuracy and speaker diarization capabilities.

Yes, all transcriptions generated through our platform can be used for commercial purposes including business meetings, podcasts, video subtitles, and content production without any additional licensing fees.