Contrastive Predictive Coding
The CPC-big model and k-means checkpoints used in Analyzing Speaker Information in Self-Supervised Models to Improve Zero-Resource Speech Processing.
Contrastive predictive coding (CPC) aims to learn representations of speech by distinguishing future observations from a set of negative examples. Previous work has shown that linear classifiers trained on CPC features can accurately predict speaker and phone labels. However, it is unclear how the features actually capture speaker and phonetic information, and whether it is possible to normalize out the irrelevant details (depending on the downstream task). In this paper, we first show that the per-utterance mean of CPC features captures speaker information to a large extent. Concretely, we find that comparing means performs well on a speaker verification task. Next, probing experiments show that standardizing the features effectively removes speaker information. Based on this observation, we propose a speaker normalization step to improve acoustic unit discovery using K-means clustering of CPC features. Finally, we show that a language model trained on the resulting units achieves some of the best results in the ZeroSpeech2021~Challenge.
Basic Usage
import torch, torchaudio
from sklearn.preprocessing import StandardScaler
# Load model checkpoints
cpc = torch.hub.load("bshall/cpc:main", "cpc").cuda()
kmeans = torch.hub.load("bshall/cpc:main", "kmeans50")
# Load audio
wav, sr = torchaudio.load("path/to/wav")
assert sr == 16000
wav = wav.unsqueeze(0).cuda()
x = cpc.encode(wav).squeeze().cpu().numpy() # Encode
x = StandardScaler().fit_transform(x) # Speaker normalize
codes = kmeans.predict(x) # Discretize
Note that the encode function is stateful (keeps the hidden state of the LSTM from previous calls).
Encode an Audio Dataset
Clone the repo and use the encode.py script:
usage: encode.py [-h] in_dir out_dir
Encode an audio dataset using CPC-big (with speaker normalization and discretization).
positional arguments:
in_dir Path to the directory to encode.
out_dir Path to the output directory.
optional arguments:
-h, --help show this help message and exit
2.2k Jan 9, 2023
11 Nov 17, 2022
5 Dec 28, 2021
29 Oct 16, 2022
8.4k Dec 26, 2022
1.5k Dec 26, 2022
5.9k Jan 3, 2023
19 Jun 30, 2022
12 Oct 26, 2022
33 Sep 22, 2022
20 Jan 9, 2023
23 Nov 30, 2022
22 Nov 13, 2022
3.2k Dec 31, 2022
1k Dec 30, 2022
1 Dec 20, 2021