ComfyUI Node
Qwen3 ASR Transcriber
A ComfyUI node in Qwen3-ASR with 10 inputs and 2 outputs.
Qwen3 ASR Transcriber
- audio
- forced_aligner
- text
- timestamps
◄model_name▾►
◄languageauto►
◄devicecuda►
◄precisionbf16►
◄max_new_tokens256►
◄flash_attention_2false►
◄chunk_size30►
◄overlap2►
CategoryQwen3-ASR
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| audio | AUDIO | The input audio to be transcribed. | |
| model_name | COMBO | The Qwen3 ASR model to use for transcription. | |
| language | COMBO | auto | The language of the audio. Set to 'auto' for automatic language detection. |
| device | COMBO | cuda | The device to run the ASR model on. |
| precision | COMBO | bf16 | The numerical precision to use for the ASR model. |
| max_new_tokens | INT | 2561–4096 | The maximum number of tokens to generate in the transcription. |
| flash_attention_2 | BOOLEAN | false | Enable Flash Attention 2 for faster inference and lower VRAM usage. |
| chunk_size | INT | 300–300 | Process audio in chunks of this many seconds. Set to 0 to disable chunking (not recommended for long audio). |
| overlap | INT | 20–10 | Overlap between chunks in seconds to maintain context. |
| forced_aligneropt | QWEN3_ALIGNER_CONF | Optional configuration for the Qwen3 Forced Aligner to generate word-level timestamps. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| text | STRING | — |
| timestamps | STRING | — |