# audio **URL:** https://discuss.pytorch.org/c/audio/9.md [Latest](https://discuss.pytorch.org/latest.md) · [Categories](https://discuss.pytorch.org/categories.md) --- ## [About the audio category](https://discuss.pytorch.org/t/about-the-audio-category/14960)
**Author:** [@apaszke](https://discuss.pytorch.org/u/apaszke)\ **Replies:** 2\ **Last updated:** [July 17, 2020, 11:16pm UTC](https://discuss.pytorch.org/t/about-the-audio-category/14960 "2020-07-17T23:16:37Z")
--- ## [Errors building torchaudio with Pytorch 2.13.0, CUDA 13.3.1, GCC 15.3.0](https://discuss.pytorch.org/t/errors-building-torchaudio-with-pytorch-2-13-0-cuda-13-3-1-gcc-15-3-0/225350)
**Author:** [@Geremia](https://discuss.pytorch.org/u/Geremia)\ **Replies:** 3\ **Last updated:** [August 31, 2026, 11:23pm UTC](https://discuss.pytorch.org/t/errors-building-torchaudio-with-pytorch-2-13-0-cuda-13-3-1-gcc-15-3-0/225350 "2026-08-31T23:23:55Z")
With Pytorch 2.13.0, CUDA 13.3.1, GCC 15.3.0; I’m getting many build errors. pip install -e . --no-deps --no-build-isolation: Defaulting to user installation because normal site-packages is not writeable Obtaining file… --- ## [Small PR for fixing CUCTC top-k shared memory offset #4198](https://discuss.pytorch.org/t/small-pr-for-fixing-cuctc-top-k-shared-memory-offset-4198/225025)
**Author:** [@lanarkite99](https://discuss.pytorch.org/u/lanarkite99)\ **Replies:** 0\ **Last updated:** [June 2, 2026, 5:11am UTC](https://discuss.pytorch.org/t/small-pr-for-fixing-cuctc-top-k-shared-memory-offset-4198/225025 "2026-06-02T05:11:35Z")
Fixes #4181 This fixes the shared-memory layout used by first\_matrix\_\_bitonic\_topk\_kernel. block\_topk\_key is a float\*, so pointer arithmetic is in units of float, not bytes. The previous code computed block\_topk\_key + … --- ## [Small torchaudio PR for non-persistent recomputable transform buffers](https://discuss.pytorch.org/t/small-torchaudio-pr-for-non-persistent-recomputable-transform-buffers/224766)
**Author:** [@Dawson](https://discuss.pytorch.org/u/Dawson)\ **Replies:** 0\ **Last updated:** [April 4, 2026, 1:39pm UTC](https://discuss.pytorch.org/t/small-torchaudio-pr-for-non-persistent-recomputable-transform-buffers/224766 "2026-04-04T13:39:02Z")
Hi everyone, I opened a small torchaudio PR addressing issue #3059 around recomputable transform buffers being persisted in state\_dict. The PR updates recomputable buffers to use register\_buffer(..., persistent=False) … --- ## [Regression with reading the CMU ARCTIC dataset](https://discuss.pytorch.org/t/regression-with-reading-the-cmu-arctic-dataset/224243)
**Author:** [@rhdunn](https://discuss.pytorch.org/u/rhdunn)\ **Replies:** 0\ **Last updated:** [December 22, 2025, 12:40pm UTC](https://discuss.pytorch.org/t/regression-with-reading-the-cmu-arctic-dataset/224243 "2025-12-22T12:40:06Z")
The commit to fix Python 3.13 support has resulted in a regression in reading the text of the CMU ARCTIC dataset for entries that contain a comma as commit Fix datasets/cmuarctic test failures when using python 3.13 (#40… --- ## [torchaudio undefined symbol](https://discuss.pytorch.org/t/torchaudio-undefined-symbol/224042)
**Author:** [@0xra0](https://discuss.pytorch.org/u/0xra0)\ **Replies:** 7\ **Last updated:** [December 15, 2025, 6:20pm UTC](https://discuss.pytorch.org/t/torchaudio-undefined-symbol/224042 "2025-12-15T18:20:36Z")
libtorchaudio.so: undefined symbol: \_ZN3c1013MessageLogger6streamB5cxx11Ev --- ## [ncclUnhandledCudaError: Call to CUDA function failed. Last error: Cuda failure 999 'unknown error](https://discuss.pytorch.org/t/ncclunhandledcudaerror-call-to-cuda-function-failed-last-error-cuda-failure-999-unknown-error/224144)
**Author:** [@so-eros](https://discuss.pytorch.org/u/so-eros)\ **Replies:** 0\ **Last updated:** [December 10, 2025, 8:09am UTC](https://discuss.pytorch.org/t/ncclunhandledcudaerror-call-to-cuda-function-failed-last-error-cuda-failure-999-unknown-error/224144 "2025-12-10T08:09:44Z")
I have installed the ZipVoice toolkit using the following set of commands:- 1.conda create -n zip python=3.13 2.git clone https://github.com/k2-fsa/ZipVoice.git 3.pip install -r requirements\_m.txt requirements\_m.txt … --- ## [Interpolate + Conv1D to Match Res Connections Shape](https://discuss.pytorch.org/t/interpolate-conv1d-to-match-res-connections-shape/211464)
**Author:** [@Inigo\_Parra](https://discuss.pytorch.org/u/Inigo_Parra)\ **Replies:** 1\ **Last updated:** [November 17, 2025, 8:35am UTC](https://discuss.pytorch.org/t/interpolate-conv1d-to-match-res-connections-shape/211464 "2025-11-17T08:35:42Z")
Hi guys, I was wondering if this forward pass is correct to align the dims of the residual connections: def forward(self, x): # print(f"Decoder input: {x.shape}") x, self\_attn = self.seq\_attention(x… --- ## [Do CTCHypothesis keep autograd history?](https://discuss.pytorch.org/t/do-ctchypothesis-keep-autograd-history/221638)
**Author:** [@alicemabille](https://discuss.pytorch.org/u/alicemabille)\ **Replies:** 0\ **Last updated:** [July 18, 2025, 12:56pm UTC](https://discuss.pytorch.org/t/do-ctchypothesis-keep-autograd-history/221638 "2025-07-18T12:56:36Z")
Hello, I would like to know if the output of CTCDecoder keeps an autograd history. I know CTCLoss does not need decoded outputs, but I would like to train my model using cross-entropy loss. Thank you. --- ## [Question about using pytorch-tcn vs. keras-tcn](https://discuss.pytorch.org/t/question-about-using-pytorch-tcn-vs-keras-tcn/220524)
**Author:** [@User9393](https://discuss.pytorch.org/u/User9393)\ **Replies:** 0\ **Last updated:** [June 2, 2025, 6:54pm UTC](https://discuss.pytorch.org/t/question-about-using-pytorch-tcn-vs-keras-tcn/220524 "2025-06-02T18:54:47Z")
Hi everyone :waving\_hand: I’m currently experimenting with Temporal Convolutional Networks (TCN) and I wanted to ask: Has anyone here used the pytorch-tcn implementation? Is it considered reliable or effective for seq… --- ## [Can't run forward pass of WaveRNN model due to unsuccessful GPU RAM allocation](https://discuss.pytorch.org/t/cant-run-forward-pass-of-wavernn-model-due-to-unsuccessful-gpu-ram-allocation/219358)
**Author:** [@j-silv](https://discuss.pytorch.org/u/j-silv)\ **Replies:** 3\ **Last updated:** [May 24, 2025, 10:16pm UTC](https://discuss.pytorch.org/t/cant-run-forward-pass-of-wavernn-model-due-to-unsuccessful-gpu-ram-allocation/219358 "2025-05-24T22:16:30Z")
When running a minimal model example of the WaveRNN, my google colab session using a T4 (16 GB RAM) GPU is unable complete a forward pass due to a cuDNN error: CUDNN\_STATUS\_NOT\_SUPPORTED. I suspect this is due to the lar… --- ## [CTC loss inputs and input lengths](https://discuss.pytorch.org/t/ctc-loss-inputs-and-input-lengths/219837)
**Author:** [@alicemabille](https://discuss.pytorch.org/u/alicemabille)\ **Replies:** 0\ **Last updated:** [May 7, 2025, 1:49pm UTC](https://discuss.pytorch.org/t/ctc-loss-inputs-and-input-lengths/219837 "2025-05-07T13:49:43Z")
Hello, I’m using CTC loss for ASR, for phoneme recognition. My model is a pretrained Whisper tiny encoder (not frozen during training) and a linear classification head as a decoder, with a LogSoftmax acitvation layer. M… --- ## [How to give muxer options to torio.io.StreamingMediaEncoder](https://discuss.pytorch.org/t/how-to-give-muxer-options-to-torio-io-streamingmediaencoder/205003)
**Author:** [@RK777](https://discuss.pytorch.org/u/RK777)\ **Replies:** 1\ **Last updated:** [April 14, 2025, 8:08am UTC](https://discuss.pytorch.org/t/how-to-give-muxer-options-to-torio-io-streamingmediaencoder/205003 "2025-04-14T08:08:17Z")
Hi, I am trying to start the StreamingMediaEncoder with format=hls and also give the following muxer options- “hls\_playlist\_type” and “hls\_time” (ffmpeg reference). But it seems that there is no argument for muxer opti… --- ## [HuBERT Pre-training for the Second Iteration without Previous Checkpoints?](https://discuss.pytorch.org/t/hubert-pre-training-for-the-second-iteration-without-previous-checkpoints/187668)
**Author:** [@jef](https://discuss.pytorch.org/u/jef)\ **Replies:** 2\ **Last updated:** [March 19, 2025, 3:39pm UTC](https://discuss.pytorch.org/t/hubert-pre-training-for-the-second-iteration-without-previous-checkpoints/187668 "2025-03-19T15:39:22Z")
I have recently completed the first round of pre-training with HuBERT (torchaudio example), using the provided learning sample, and have finished preprocessing for the second iteration. I have a question regarding the pr… --- ## [How to Export \`torchaudio.models.decoder.ctc\_decoder\` to TorchScript for C++ Deployment?](https://discuss.pytorch.org/t/how-to-export-torchaudio-models-decoder-ctc-decoder-to-torchscript-for-c-deployment/217774)
**Author:** [@mariaalfaroc](https://discuss.pytorch.org/u/mariaalfaroc)\ **Replies:** 1\ **Last updated:** [March 18, 2025, 10:53am UTC](https://discuss.pytorch.org/t/how-to-export-torchaudio-models-decoder-ctc-decoder-to-torchscript-for-c-deployment/217774 "2025-03-18T10:53:52Z")
Hello everyone, I’m trying to export torchaudio.models.decoder.ctc\_decoder to TorchScript so that I can use it in a C++ application. However, I have encountered several issues along the way. What I Tried Using torch.… --- ## [\[PR\] Torchaudio incompatible with python flag -OO due to \_\_doc\_\_ being None](https://discuss.pytorch.org/t/pr-torchaudio-incompatible-with-python-flag-oo-due-to-doc-being-none/216897)
**Author:** [@FremyCompany](https://discuss.pytorch.org/u/FremyCompany)\ **Replies:** 2\ **Last updated:** [February 19, 2025, 3:16pm UTC](https://discuss.pytorch.org/t/pr-torchaudio-incompatible-with-python-flag-oo-due-to-doc-being-none/216897 "2025-02-19T15:16:06Z")
I opened the follwing pull request to fix the issue I think I’m facing. Avoid crash when doc is optimized away to None (-OO) by FremyCompany · Pull Request #3884 · pytorch/audio Could someone take a look? --- ## [How to use filtfilt() function?](https://discuss.pytorch.org/t/how-to-use-filtfilt-function/216443)
**Author:** [@elinliu0823](https://discuss.pytorch.org/u/elinliu0823)\ **Replies:** 1\ **Last updated:** [February 10, 2025, 8:55am UTC](https://discuss.pytorch.org/t/how-to-use-filtfilt-function/216443 "2025-02-10T08:55:11Z")
I’m trying to move from scipy to torchaudio. Here is my code below: from torchaudio.functional.filtering import filtfilt from scipy import signal bh, ah = signal.butter(N=5, Wn=48, btype="high", fs=16000) audio = sam… --- ## [Issues with CTCLoss](https://discuss.pytorch.org/t/issues-with-ctcloss/216390)
**Author:** [@Butzti](https://discuss.pytorch.org/u/Butzti)\ **Replies:** 0\ **Last updated:** [February 8, 2025, 1:34am UTC](https://discuss.pytorch.org/t/issues-with-ctcloss/216390 "2025-02-08T01:34:34Z")
Hey, So I’m trying to implement the DeepSpeech2 architecture using PyTorch but I’m currently having trouble with the CTCLoss. The documentation says, that the input to the CTCLoss should be logarithmic probabilites for … --- ## [Torchaudio is not able to access FFmpeg](https://discuss.pytorch.org/t/torchaudio-is-not-able-to-access-ffmpeg/202949)
**Author:** [@Dylen](https://discuss.pytorch.org/u/Dylen)\ **Replies:** 5\ **Last updated:** [January 20, 2025, 5:33pm UTC](https://discuss.pytorch.org/t/torchaudio-is-not-able-to-access-ffmpeg/202949 "2025-01-20T17:33:29Z")
I am new to torchaudio, and I am trying to use StreamReader to get data from my microphone, but I cannot get it to start because it can’t access FFmpeg. from torchaudio.io import StreamReader streamer = StreamReader( … --- ## [Siamese Network for gender recognition](https://discuss.pytorch.org/t/siamese-network-for-gender-recognition/57208)
**Author:** [@dengio](https://discuss.pytorch.org/u/dengio)\ **Replies:** 2\ **Last updated:** [January 14, 2025, 1:41am UTC](https://discuss.pytorch.org/t/siamese-network-for-gender-recognition/57208 "2025-01-14T01:41:45Z")
Hi everyone, recently I started to read about Siamese Nets and I wanted to try this type of model on a gender recognition task. My data is a .csv dataset containing ~3000 n-vectors of audio features (n=20). The loss func… --- ## [Please Help: \`torchaudio\` does not find the \`sox\` backend on macOS](https://discuss.pytorch.org/t/please-help-torchaudio-does-not-find-the-sox-backend-on-macos/214313)
**Author:** [@ThomasHezard](https://discuss.pytorch.org/u/ThomasHezard)\ **Replies:** 3\ **Last updated:** [January 5, 2025, 8:12pm UTC](https://discuss.pytorch.org/t/please-help-torchaudio-does-not-find-the-sox-backend-on-macos/214313 "2025-01-05T20:12:12Z")
Hello, I’ve been struggling with torchaudio since I migrated to my new macBook Pro. The problem is the following: torchaudio can not initialize the sox backend when the Python environment is managed by uv, and I debu… --- ## [Understanding channel dim in audio processing](https://discuss.pytorch.org/t/understanding-channel-dim-in-audio-processing/212777)
**Author:** [@axion66](https://discuss.pytorch.org/u/axion66)\ **Replies:** 1\ **Last updated:** [November 26, 2024, 7:16am UTC](https://discuss.pytorch.org/t/understanding-channel-dim-in-audio-processing/212777 "2024-11-26T07:16:03Z")
Hi! I’m working on generative audio model, and I want to what’s the most common way of selecting channel dimension. I saw some people after doing torch.stft or spectrogram, add additional dimension to have (batch,1,freq… --- ## [Stft behavior with center=False and n\_fft \> win\_length](https://discuss.pytorch.org/t/stft-behavior-with-center-false-and-n-fft-win-length/213169)
**Author:** [@nedlrichards](https://discuss.pytorch.org/u/nedlrichards)\ **Replies:** 0\ **Last updated:** [November 19, 2024, 8:21pm UTC](https://discuss.pytorch.org/t/stft-behavior-with-center-false-and-n-fft-win-length/213169 "2024-11-19T20:21:19Z")
I am implimenting a stft library that uses torch under the hood but uses the parameter conventions of scipy’s ShortTimeFFT. I am having trouble understanding the torch behavior when the stft fft is zero padded and center… --- ## [Torchaudio.functional.speed very slow when called repeatedly using randomly generated factors](https://discuss.pytorch.org/t/torchaudio-functional-speed-very-slow-when-called-repeatedly-using-randomly-generated-factors/212781)
**Author:** [@MikeK](https://discuss.pytorch.org/u/MikeK)\ **Replies:** 0\ **Last updated:** [November 11, 2024, 1:35am UTC](https://discuss.pytorch.org/t/torchaudio-functional-speed-very-slow-when-called-repeatedly-using-randomly-generated-factors/212781 "2024-11-11T01:35:41Z")
Hello, I’m trying to write a function that applies random augmentations to audio files, which has been converted to pytorch tensors in a prior operation. The problem is, the code I wrote runs really slow, I have located … --- ## [How to align the resample result of librosa and torchaudio](https://discuss.pytorch.org/t/how-to-align-the-resample-result-of-librosa-and-torchaudio/209447)
**Author:** [@wadewang](https://discuss.pytorch.org/u/wadewang)\ **Replies:** 2\ **Last updated:** [October 31, 2024, 2:32am UTC](https://discuss.pytorch.org/t/how-to-align-the-resample-result-of-librosa-and-torchaudio/209447 "2024-10-31T02:32:07Z")
The speed of torchauido resample attracts me after compare against librosa. But I found their results are a little different. I have read this documentation: Audio Resampling — Torchaudio 2.4.0 documentation , but it ju… --- ## [Audios wav pytorch](https://discuss.pytorch.org/t/audios-wav-pytorch/211673)
**Author:** [@jsanchez](https://discuss.pytorch.org/u/jsanchez)\ **Replies:** 3\ **Last updated:** [October 16, 2024, 4:16pm UTC](https://discuss.pytorch.org/t/audios-wav-pytorch/211673 "2024-10-16T16:16:06Z")
good day your support this error RuntimeError: mat1 dim 1 must match mat2 dim 0, I am working pytorch to classify wav audios, BATCH\_SIZE = 32 ###128 para 4conv EPOCHS = 10 LEARNING\_RATE = 0.001 ANNOTATIONS\_FILE = “…/… --- ## [Transposed Convolutions](https://discuss.pytorch.org/t/transposed-convolutions/211366)
**Author:** [@Inigo\_Parra](https://discuss.pytorch.org/u/Inigo_Parra)\ **Replies:** 2\ **Last updated:** [October 12, 2024, 4:14am UTC](https://discuss.pytorch.org/t/transposed-convolutions/211366 "2024-10-12T04:14:23Z")
Hi! I am having a hard time understanding the ConvTranspose1D layer. I would really appreciate if someone could help me understand the transposed convolutions better (in detail if possible) Thank you! --- ## [Frequencies and time arays for spectrogram](https://discuss.pytorch.org/t/frequencies-and-time-arays-for-spectrogram/209438)
**Author:** [@sammlapp](https://discuss.pytorch.org/u/sammlapp)\ **Replies:** 3\ **Last updated:** [October 8, 2024, 8:41pm UTC](https://discuss.pytorch.org/t/frequencies-and-time-arays-for-spectrogram/209438 "2024-10-08T20:41:36Z")
When we use the torchaudio.transforms.spectrogram transform we get a 2d array with dimensions corresponding to frequency and time. The exact frequencies and times of rows and columns depend on the spectrogram parameters. … --- ## [Wav2Vec2FABundle emission frame size](https://discuss.pytorch.org/t/wav2vec2fabundle-emission-frame-size/211311)
**Author:** [@pytorque](https://discuss.pytorch.org/u/pytorque)\ **Replies:** 1\ **Last updated:** [October 8, 2024, 5:45pm UTC](https://discuss.pytorch.org/t/wav2vec2fabundle-emission-frame-size/211311 "2024-10-08T17:45:04Z")
The example waveform in the forced alignment tutorial has 54k frames, and is sampled in 16 kHz. It renders an emission tensor of size \[1, 169, 28\] if I remember correctly. There are 28 labels, and seemingly 169 binned (?… --- ## [Torchaudio: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!](https://discuss.pytorch.org/t/torchaudio-expected-all-tensors-to-be-on-the-same-device-but-found-at-least-two-devices-cuda-0-and-cpu/98445)
**Author:** [@Mahmoud\_Abdelkhalek](https://discuss.pytorch.org/u/Mahmoud_Abdelkhalek)\ **Replies:** 3\ **Last updated:** [October 2, 2024, 12:03pm UTC](https://discuss.pytorch.org/t/torchaudio-expected-all-tensors-to-be-on-the-same-device-but-found-at-least-two-devices-cuda-0-and-cpu/98445 "2024-10-02T12:03:57Z")
Here is my code: import sys import torch import torchaudio def train(net,dataloader,loss\_func,optimizer,device): # put in training mode net.train() # to compute training accuracy num… [Next page](https://discuss.pytorch.org/c/audio/9.md?page=1)