Preprint
Conformer: Convolution-augmented Transformer for Speech Recognition
Anmol Gulati et al., arXiv (Interspeech 2020), 2020
Read the originalRead on October 16, 2026
- What it is
- Google's paper proposing the Conformer encoder, which combines convolution layers with transformer self-attention, and reporting LibriSpeech word error rates with and without an external language model.
- How much weight it carries
- Submitted to Interspeech 2020 and now the basis of most production speech encoders. Numbers are from the LibriSpeech benchmark, which is read audiobook speech.
Facts taken from this source
- The Conformer paper introduces the convolution-augmented transformer for speech recognition, named Conformer, which combines convolution neural networks with transformers.
“we propose the convolution-augmented transformer for speech recognition, named Conformer” (Abstract)
- The Conformer model reached word error rates of 2.1 percent on LibriSpeech test-clean and 4.3 percent on test-other without a language model, and 1.9 and 3.9 percent with one.
“our model achieves WER of 2.1%/4.3% without using a language model and 1.9%/3.9% with an external language model on test/testother” (Abstract)
Pages that rest on this source
- Speech Recognition Models: How Neural ASR WorksGuide
- Which neural network is best for speech recognition?Question
- ConformerGlossary
- Speech Recognition: What It Is and How Well It WorksTopic
- How Speech Recognition Works, From Sound to TextGuide
- Is Speech to Text AI? What Kind of AI Voice Typing UsesGuide
- Word Error Rate (WER): How to Calculate and Read ItGuide
- Speech Recognition Benchmarks and LeaderboardsGuide