Audio

Conformer block

A transformer block with a convolution module inside. Attention for global context, convs for local acoustic detail.

First page of Conformer: Convolution-augmented Transformer for Speech RecognitionIntroduced inMay 2020Conformer: Convolution-augmented Transformer for Speech RecognitionGulati et al. · arXiv 2005.08100 ↗
Conformer: attention and convolution, togetherattention hears the whole utterance · the conv module keeps the local detailfeed-forward (½)self-attention · globalconvolution · local detailfeed-forward (½)

The Conformer sandwiches a depthwise-convolution module between attention and feed-forward layers (in a macaron FFN-attention-conv-FFN pattern), usually with relative position encoding in the attention. Convolutions capture local spectral patterns that pure attention handles poorly, which made Conformers the dominant speech-encoder architecture (NVIDIA’s FastConformer here).

Adoption over time

Share of new models that have included Conformer blocks over time.

1%202220232024202520261%20222023202420252026

See it in real models

Open any of these on hfviewer to find this block in the interactive architecture graph.

ResembleAI/chatterbox architecture graphResembleAI/chatterboxtext-to-speech · ↓ 2.0M · ♡ 2kOpen in visualizer nvidia/nemotron-3.5-asr-streaming-0.6b architecture graphnvidia/nemotron-3.5-asr-streaming-0.6bautomatic-speech-recognition · ↓ 1.0M · ♡ 1kOpen in visualizer nvidia/parakeet-tdt-0.6b-v3 architecture graphnvidia/parakeet-tdt-0.6b-v3automatic-speech-recognition · ↓ 725k · ♡ 1kOpen in visualizer CohereLabs/cohere-transcribe-03-2026 architecture graphCohereLabs/cohere-transcribe-03-2026automatic-speech-recognition · ↓ 531k · ♡ 1kOpen in visualizer facebook/seamless-m4t-v2-large architecture graphfacebook/seamless-m4t-v2-largeautomatic-speech-recognition · ↓ 337k · ♡ 1kOpen in visualizer google/gemma-3n-E2B-it architecture graphgoogle/gemma-3n-E2B-itimage-text-to-text · ↓ 292k · ♡ 320Open in visualizer ibm-granite/granite-speech-4.1-2b architecture graphibm-granite/granite-speech-4.1-2bautomatic-speech-recognition · ↓ 247k · ♡ 157Open in visualizer ai-sage/GigaAM-v3 architecture graphai-sage/GigaAM-v3automatic-speech-recognition · ↓ 238k · ♡ 141Open in visualizer nvidia/nemotron-speech-streaming-en-0.6b architecture graphnvidia/nemotron-speech-streaming-en-0.6bautomatic-speech-recognition · ↓ 143k · ♡ 612Open in visualizer nvidia/diar_streaming_sortformer_4spk-v2 architecture graphnvidia/diar_streaming_sortformer_4spk-v2automatic-speech-recognition · ↓ 31k · ♡ 139Open in visualizer nvidia/canary-qwen-2.5b architecture graphnvidia/canary-qwen-2.5bautomatic-speech-recognition · ↓ 28k · ♡ 454Open in visualizer nvidia/parakeet-tdt_ctc-0.6b-ja architecture graphnvidia/parakeet-tdt_ctc-0.6b-jaautomatic-speech-recognition · ↓ 5k · ♡ 60Open in visualizer

Browse all 32 models with this in the catalog →

Related concepts

Mel spectrogram front-end Convolutional subsampling RVQ codebooks

hfviewer renders the full architecture of 4,000+ Hugging Face models as interactive graphs. Hover any block to see what it does, with this model’s real numbers.

Browse all model graphs →