Scaling Up Online Speech Recognition Using ConvNets
By Vineel Pratap, Qiantong Xu, Jacob Kahn, Gilad Avidov, Tatiana Likhomanenko, Awni Y. Hannun, Vitaliy Liptchinsky, Gabriel Synnaeve, Ronan Collobert. 2020
48 citations recorded by the source corpus.
We design an online end-to-end speech recognition system based on Time-Depth Separable (TDS) convolutions and Connectionist Temporal Classification (CTC). We improve the core TDS architecture in order to limit the future context and hence reduce latency while maintaining accuracy. The system has almost three times the throughput of a well tuned hybrid ASR baseline while also having lower latency and a better word error rate. Also important to the efficiency of the recognizer is our highly optimized beam search decoder. To show the impact of our design choices, we analyze throughput, latency, accuracy, and discuss how these metrics can be tuned based on the user requirements.
Subfields: Artificial Intelligence, Speech & Audio, Signal Processing
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Gabriel Synnaeve
- Ronan Collobert
- Qiantong Xu
- Vineel Pratap
- Michael Auli
- Anuroop Sriram
- Alexei Baevski
- Wei-Ning Hsu
In Speech & Audio
- Gabriel Synnaeve
- Ronan Collobert
- Qiantong Xu
- Vineel Pratap
- Anuroop Sriram
- Alexei Baevski
- Michael Auli
- Wei-Ning Hsu
In Signal Processing
- Gabriel Synnaeve
- Ronan Collobert
- Qiantong Xu
- Vineel Pratap
- Anuroop Sriram
- Michael Auli
- Wei-Ning Hsu
- Alexei Baevski
Citations in the corpus
- MLS: A Large-Scale Multilingual Dataset for Speech Research (2020, methodology): “Since the audio files can be very long (>30 minutes), we have used wav2letter@anywhere framework [10] to perform the inference in a streaming fashion.”