Cohere Transcribe

This section describes how to use models from https://huggingface.co/CohereLabs/cohere-transcribe-03-2026.

A single model from Cohere Transcribe supports the following 14 languages:

• European: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish

• AIPAC: Chinese (Mandarin), Japanese, Korean, Vietnamese

• MENA: Arabic

The converted sherpa-onnx model is an offline non-streaming ASR model. When decoding, you need to specify the input language, for example en for English, zh for Chinese, and de for German.

We have provided Cohere Transcribe examples for the following 12 programming languages:

C++	C	Python	C#
Go	Kotlin	Java	JavaScript (Node.js and node-addon)
Swift	Dart (Support Flutter)	Rust	Object Pascal

In the following, we describe how to download a pre-trained model and where to find the corresponding examples in k2-fsa/sherpa-onnx.

• Pre-trained model
  • Decode a short audio file
  • Decode a long audio file with VAD (Example 1/2, English)
  • Decode a long audio file with VAD (Example 2/2, Chinese)
• Huggingface space
• Examples for 12 programming languages
  • C++
  • C
  • Python
  • C#
  • Go
  • Kotlin
  • Java
  • JavaScript
  • Swift
  • Dart
  • Rust
  • Object Pascal