Acoustic Modelling from the Signal Domain Using CNNs

By Pegah Ghahremani, Vimal Manohar, Daniel Povey, Sanjeev Khudanpur. 2016

92 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Model Architectures, 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

In Model Architectures

In Speech & Audio

In Signal Processing

Citations in the corpus

CHiME-6 Challenge: Tackling Multispeaker Speech Recognition for Unsegmented Recordings (2020, methodology): “As a neural network architecture, we use 40-dimensional MFCC features as input, 5 TDNN layers, and 2 layers of statistics pooling [46].”
Deep Speaker: an End-to-End Neural Speaker Embedding System (2017, methodology): “[6] used the last frame output of a long short-term memory (LSTM) [9] model as an utterance-level speaker embedding, while [7] used a network-in-network (NIN) [10] nonlinearity followed by an utterance-level pooling layer to aggregate frame-level representations.”
TaSNet: Time-Domain Audio Separation Network for Real-Time, Single-Channel Speech Separation (2017, methodology): “In recent years, this approach has been used successfully applied in tasks such as speech recognition, speech synthesis and speech enhancement [9, 10, 11, 12, 13], but waveform-level speech separation with deep learning has not been investigated yet.”
Kaleidoscope: An Efficient, Learnable Representation For All Structured Linear Maps (2020, background): “Other approaches that can learn from raw audio can be found in (Palaz et al., 2013; Collobert et al., 2016; Ghahremani et al., 2016).”
Speaker Recognition from Raw Waveform with SincNet (2018, background): “To mitigate this drawback, some recent works have proposed directly feeding the network with spectrogram bins [19–21] or even with raw waveforms [22–34].”
Attention-based Wav2Text with feature transfer learning (2017, methodology): “[5] recently proposed a CNN time-delay neural network (CNN-TDNN) with network-innetwork (NIN) architecture, and also showed that their model outperformed MFCC-based TDNN on the Wall Street Journal (WSJ) [6] task.”
Deep Learning for Audio Signal Processing (2019, background): “However, some works report improvements using raw waveforms for analysis tasks [25], [146], [147], and some attempt to find a way in between by designing and/or initializing the first layers of a deep learning system to mimic engineered”
Deep neural network-based speaker embeddings for end-to-end speaker verification (2016, background): “The network activations are a type of network-in-network (NIN) nonlinearity recently introduced in [26].”
AISHELL-4: An Open Source Dataset for Speech Enhancement, Separation, Recognition and Speaker Diarization in Conference Scenario (2021, methodology): “The diarizaiton module includes three components: SAD, speaker embedding extractor and clustering.”
The STC System for the CHiME-6 Challenge (2020, methodology): “To perform multi-stream self-attention layers, we replaced 8-layer TDNN-F from the CNN-TDNN-F described above, with 5 streams of 7 TDNN-F layers and 12 self-attention heads per stream.”
Interpretable Convolutional Filters with SincNet (2018, background): “employed in many state-of-the-art systems [19, 20, 21], directly feeding a CNN with spectrogram bins [22, 23, 24] or even with raw audio samples [25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37] is an approach of increasing popularity.”