MUSAN: A Music, Speech, and Noise Corpus
By David Snyder, Guoguo Chen, Daniel Povey. 2015
1,702 citations recorded by the source corpus.
This report introduces a new corpus of music, speech, and noise. This dataset is suitable for training models for voice activity detection (VAD) and music/speech discrimination. Our corpus is released under a flexible Creative Commons license. The dataset consists of music from several genres, speech from twelve languages, and a wide assortment of technical and non-technical noises. We demonstrate use of this corpus for music/speech discrimination on Broadcast news and VAD for speaker identification.
Subfields: Artificial Intelligence, Evaluation & Benchmarks, Speech & Audio
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 Evaluation & Benchmarks
- Shinji Watanabe
- Gabriel Synnaeve
- Hung-yi Lee
- Emmanuel Dupoux
- Jinyu Li
- Zhuo Chen
- Morgane Rivière
- Abdel-rahman Mohamed
In Speech & Audio
- Shinji Watanabe
- Jinyu Li
- Wei-Ning Hsu
- Abdel-rahman Mohamed
- Daniel Povey
- Emmanuel Dupoux
- Sanjeev Khudanpur
- Gabriel Synnaeve
Citations in the corpus
- Earnings-21: A Practical Benchmark for ASR in the Wild (2021, background): “Whereas there exist multiple techniques that allow to adapt an ASR system to various acoustic conditions [1, 2, 3], it is also necessary to evaluate the system in its target operating conditions.”
- Privacy-Preserving Adversarial Representation Learning in ASR: Reality or Illusion? (2019, positive): “The train-960 subset was augmented using room impulse responses, isotropic and point-source noises [34] as well as music and speech [35] as per the standard sre16 recipe for training x-vectors [24] from the Kaldi toolkit [36], which we adapted to Librispeech.”
- SpeechX: Neural Codec Language Model as a Versatile Speech Transformer (2023, methodology): “Noise suppression: We mixed each test sample with a noise sample randomly picked from the MUSAN dataset [34] at a signal-to-noise ratio (SNR) which was randomly determined from the range between 0 dB and 20 dB.”
- WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing (2021, methodology): “For data pre-processing, we apply online data augmentation using the MUSAN [54] noise, DNS noise [48] and the RIR 4 reverberation with probability 0.”
- Data Augmenting Contrastive Learning of Speech Representations in the Time Domain (2020, methodology): “We therefore explored systematically the effect of the spectral characteristics of noise by filtering sounds from the MUSAN dataset [25] in successive frequency bands.”
- Pyannote.Audio: Neural Building Blocks for Speaker Diarization (2019, methodology): “For instance, it supports extracting features from random audio chunks while applying additive noise from databases such as MUSAN [11].”
- End-to-End Speaker Diarization for an Unknown Number of Speakers with Encoder-Decoder Based Attractors (2020, methodology): “For the training and evaluation, we used simulated mixtures created from Switchboard-2 (Phase I & II & III), Switchboard Cellular (Part 1 & 2), and the NIST Speaker Recognition Evaluation (2004 & 2005 & 2006 & 2008) for speech and the MUSAN corpus [28] for noise with simulated room impulse responses”
- Limitations of Weak Labels for Embedding and Tagging (2020, methodology): “We created 10 s sound clips using foreground sound events from the Freesound dataset [25,26] and backgrounds from SINS [27] and MUSAN [28].”
- Semi-supervised Learning for Multi-speaker Text-to-speech Synthesis Using Discrete Speech Representation (2020, methodology): “The noises (10-30dB SNR) used in VCTK(n) are randomly selected from the MUSAN dataset [31].”
- Wespeaker: A Research and Production Oriented Speaker Embedding Learning Toolkit (2022, methodology): “The audios from the MUSAN dataset [30] are used as additive noises, while the simulated room impulse responses (RIRs)(7) are used for the reverberation.”
- A study on data augmentation of reverberant speech for robust speech recognition (2017, methodology): “The point-source noises are sampled from the Freesound portion of the MUSAN corpus [13].”
- The SpeakIn System for VoxCeleb Speaker Recognition Challange 2021 (2021, methodology): “Both foreground and background noises are added, and they are randomly selected from MUSAN and RIRs noises.”
- Is Someone Speaking?: Exploring Long-term Temporal Features for Audio-visual Active Speaker Detection (2021, positive): “As the Columbia ASD dataset is an open dataset, we apply the additional sources from RIRs data [25] and the MUSAN dataset [42] to perform audio augmentation on the TalkSet during training, and evaluate the performance using Sklearn library5.”
- ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification (2020, methodology): “The first set of augmentations follow the KALDI recipe [12] in combination with the publicly available MUSAN dataset (babble, noise) [18] and the RIR dataset (reverb) provided in [19].”
- Multimodal Self-Supervised Learning of General Audio Representations (2021, positive): “MUSAN [34] is used for detection of music, speech, or noise.”
- Self-supervised audio representation learning for mobile devices (2019, positive): “The Spoken Language Identification dataset [30] contains samples that belong to three different languages: English, Spanish and German, while the MUSAN dataset [38] distinguishes across three classes, namely music, speech and noise.”
- Efficient Keyword Spotting Using Dilated Convolutions and Gating (2018, background): “Noisy conditions are simulated by augmenting samples with music and noise background audio from Musan [13].”
- Voices Obscured in Complex Environmental Settings (VOICES) corpus (2018, methodology): “Babble noise was constructed using the ”us-gov” subset of the MUSAN corpus[10].”