Nova-3 General

Deepgram Nova-3 General extends the Nova-3 architecture with multilingual capabilities, supporting 30+ languages with automatic language detection. It is designed for global voice applications, multilingual contact centers, international content pipelines, and cross-language analytics — all without requiring pre-specification of the input language.

Technical Specifications

Performance Benchmarks

• Supports 30+ languages with competitive per-language word error rates.
• Automatic language detection runs in parallel with transcription.
• Sub-second latency in streaming mode.
• Handles code-switching in select language pairs.
• Consistent accuracy across diverse accents and recording conditions.

Architecture Breakdown

Nova-3 General shares the same end-to-end deep learning foundation as Nova-3, extended with a multilingual language head trained on diverse language corpora. Language identification is embedded in the transcription pipeline, removing the need for a separate detection step.

Pricing

• $0.01001 / min

Core Features & Capabilities

• Multilingual Streaming: Real-time transcription across 30+ languages via WebSocket.
• Auto Language Detection: Dynamically identifies spoken language — no pre-configuration needed.
• Speaker Diarization: Labels individual speakers across multilingual audio sessions.
• Smart Formatting: Locale-aware number, date, and punctuation formatting.
• Intent & Topic Detection: Custom and model-detected intents and topics across languages.
• Entity Detection: Extracts key entities from multilingual audio content.
• Custom Vocabulary (keyterm): Add domain-specific terms per language to improve accuracy.
• Utterance Segmentation: Segments multilingual streams into labeled speech units.

Comparison with Other Models

• VS Deepgram Nova-3: Nova-3 General adds multilingual support at the same price; Nova-3 offers maximum accuracy for English-only use cases.
• VS Deepgram Nova-3 Medical: Nova-3 General is the general-purpose multilingual model; Nova-3 Medical is specialized for healthcare audio at a lower price per minute.

‍