Signal estimation from modified short-time Fourier transform

By Daniel W. Griffin, Jae S. Lim. 1983

Subfields: 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 Speech & Audio

1. Wei-Ning Hsu
2. Xu Tan
3. Hung-yi Lee
4. Yu Zhang
5. Junichi Yamagishi
6. Tao Qin
7. Jinyu Li
8. Tie-Yan Liu

In Signal Processing

1. Junichi Yamagishi
2. Xu Tan
3. Wei-Ning Hsu
4. Tao Qin
5. Jinyu Li
6. Frank K. Soong
7. Hung-yi Lee
8. Tie-Yan Liu

Citations in the corpus

• Waveglow: A Flow-based Generative Network for Speech Synthesis (2018, methodology): “As baseline for mean opinion score we compare the popular Griffin-Lim algorithm [16].”
• Towards Unsupervised Speech Recognition and Synthesis with Quantized Speech Representation Learning (2019, methodology): “Griffin-Lim algorithm [22] is used to convert spectrogram to waveform and adapting a vocoder is left as future work.”
• VQVAE Unsupervised Unit Discovery and Multi-scale Code2Spec Inverter for Zerospeech Challenge 2019 (2019, methodology): “given the predicted linear magnitude spectrogram M̂ , we reconstruct the missing phase spectrogram with the Griffin-Lim algorithm [20] and applied the inverse short-term Fourier transform (STFT) to generate the waveform.”
• MERLOT RESERVE: Neural Script Knowledge through Vision and Language and Sound (2022, methodology): “We used a small number of mel spectrogram bins (64) because we found that at that threshold, we could reconstruct the original sequence at an acceptable level using the Griffin-Lim algorithm, [52] which itself might be a lower bound on quality as neural methods trained for this purpose have been sho”
• Natural TTS Synthesis by Conditioning Wavenet on MEL Spectrogram Predictions (2017, methodology): “We also compare to the original Tacotron that predicts linear spectrograms and uses Griffin-Lim to synthesize audio, as well as concatenative [30] and parametric [31] baseline systems, both of which have been used in production at Google.”
• Tacotron: Towards End-to-End Speech Synthesis (2017, methodology): “We use the Griffin-Lim algorithm [21] to synthesize waveform from the predicted spectrogram.”
• Neural Speech Synthesis with Transformer Network (2018, methodology): “(2017), is an end-to-end generative text-tospeech model, which can be trained by 〈text, spectrogram〉 pairs directly from scratch, and synthesizes speech audios with generated spectrograms by Griffin Lim algorithm (Griffin and Lim 1984).”
• Semi-Supervised Neural Architecture Search (2020, methodology): “Using Griffin-Lim as the vocoder to synthesize the speech, the ground-truth mel-spectrograms achieves 3.26 MOS, Transformer TTS achieves 2.25, NAO achieves 2.60 and SemiNAS achieves 2.66.”
• UniAudio: An Audio Foundation Model Toward Universal Audio Generation (2023, background): “…al., 2021; Shen et al., 2023); (3) their designs are inspired by and derived from linguistics and phonetics (Zen et al., 2013), signal processing (Griffin & Lim, 1984), auditory perception (Shadle & Damper, 2001) and machine learning (Wang et al., 2016) research etc; (4) they use different…”
• Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis (2023, methodology): “Traditional methods usually rely on the Griffin-Lim algorithm (Griffin and Lim, 1984), which iteratively estimate the phase by enforcing spectrogram consistency.”
• A Survey on Neural Speech Synthesis (2021, methodology): “RNN-based Models (e.g., Tacotron Series) Tacotron [376] leverages an encoder-attentiondecoder framework and takes characters as input9 and outputs linear-spectrograms, and uses GriffinLim algorithm [92] to generate waveform.”
• Speech Resynthesis from Discrete Disentangled Self-Supervised Representations (2021, positive): “Next, a vocoder is applied to reconstruct the phase from the Mel-spectrogram (e.g., Griffin-Lim [6], WaveNet [7], WaveGlow [8], or HiFi-GAN [9]).”
• Semi-supervised Training for Improving Data Efficiency in End-to-end Speech Synthesis (2018, methodology): “Using Griffin-Lim allows much faster experiment cycles.”
• Towards End-to-End Prosody Transfer for Expressive Speech Synthesis with Tacotron (2018, methodology): “The predicted mel spectrograms can either be synthesized directly to the timedomain via a WaveNet vocoder (Shen et al., 2017), or by first learning a linear spectrogram prediction network, and then applying Griffin-Lim spectrogram inversion (Griffin & Lim, 1984).”
• Deep Voice 3: 2000-Speaker Neural Text-to-Speech (2017, methodology): “L1 loss on linear-scale (log-magnitude) spectrograms for use with Griffin-Lim, 2.”
• ClariNet: Parallel Wave Generation in End-to-End Text-to-Speech (2018, methodology): “, 2016), the Griffin-Lim algorithm (Griffin and Lim, 1984), or a separately trained neural vocoder (Ping et al.”
• Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation (2018, methodology): “Although methods for phase reconstruction can be applied to alleviate this issue [11], [13], [14], the performance of the method remains suboptimal.”