Learning Joint Acoustic-Phonetic Word Embeddings

Mohamed El-Geish

REVIEW 4 major objections 6 minor 47 references

Learning Joint Acoustic-Phonetic Word Embeddings

T0 review · 4 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that acoustic and phonetic word forms can be embedded into one shared vector space, trained only from weak binary labels, with a best test F1 of 0.95.

desk verdict An interesting method for acoustic-phonetic embeddings with a useful semi-supervised hard-negative mining trick, but the headline F1=0.95 is likely inflated by example-level train/test leakage; needs a re-run before the result is credible. read the letter →

arxiv 1908.00493 v1 pith:UVP52RCN submitted 2019-08-01 cs.LG cs.CLcs.SDeess.ASstat.ML

Mohamed El-Geish This is my paper · ORCID

classification cs.LGcs.CLcs.SDeess.ASstat.ML

keywords acousticwordembeddingsphoneticSiameseconvolutionalnetworkweaksupervisioncontrastivelosshardnegativeminingself-labelingspeechrecognition

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to learn a single vector space in which any word, whether it arrives as audio or as a phonetic spelling, becomes a fixed-length vector, and the distance between two vectors reflects how similar the words sound. To avoid expensive human pairwise-distance labels, the author frames the task as weakly supervised binary classification: decide whether an acoustic example and a phonetic example represent the same word. The central empirical claim is that a Siamese convolutional network trained with contrastive loss and aggressive hard-negative mining reaches an F1 score of 0.95 on this binary task at a distance threshold of 0.5. If true, speech systems could compare words across modalities directly, without first converting audio to text, which is valuable for keyword spotting, query-by-example search, and ASR hypothesis reranking.

What carries the argument

The load-bearing mechanism is a Siamese neural network with two encoders, $f(x_a)$ and $g(x_p)$, that map acoustic and phonetic inputs to $\ell^2$-normalized embeddings, scored by a distance $D(f(x_a),g(x_p))$ and trained with the contrastive loss $\mathcal{L}=\frac{1}{N}\sum_i[(1-y^{(i)})(D^{(i)})^2 + y^{(i)}\max(0,m-D^{(i)})^2]$. The encoders mirror each other architecturally, except the acoustic side gets input dropout; phonetic input is a one-hot matrix of phones, and audio is a mel-spectrogram tensor. What carries the argument is the combination of that loss with three hard-negative mining techniques: ASR substitution pairs, synthetic pairs grouped by phonetic-edit distance, and self-labeling using a k-d tree over phonetic embeddings. These mining techniques convert weak supervision into a training signal strong enough for the model to learn sound-based closeness.

What would settle it

Take a randomly sampled set of word pairs from the same recordings, have human listeners judge whether each pair sounds like the same word, and compare those judgments with both the pipeline-generated labels and the model's predicted labels; if human agreement with either is far below the reported 0.95 F1, the result is an artifact of the labeling pipeline rather than a learned property of acoustic-phonetic similarity.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that joint acoustic-phonetic word embeddings can be learned without ground-truth word-sound distances, using only weak binary labels derived from an automatic pipeline. The pipeline force-aligns noisy human transcripts with ensemble ASR hypotheses, labels aligned pairs as similar, labels ASR substitution errors as hard negatives, synthesizes additional hard negatives by phonetic-edit distance, and periodically self-labels new hard negatives from a k-dimensional tree over the training lexicon's phonetic embeddings. Under these labels, the best model—a Siamese CNN with two convolutional blocks, pooling, two 512-unit dense layers, dropout, a 512-dimensional embedding, and cosine distance—achieves test F1 0.95. The author further reports that t-SNE projections show word clusters and sound-based analogies, while manual inspection of mistakes points to noisy audio, far-field speech, and accents.

Load-bearing premise

The load-bearing premise is that the automatically generated pair labels are correct enough: aligned transcript-and-audio pairs genuinely sound alike, and substitution-derived negative pairs genuinely sound different.

Editorial extensions

If this is right

A query audio word can be compared directly with phonetic spellings in a fixed-dimensional space, which makes query-by-example search and keyword spotting simpler because no ASR transcript is needed for the query.
Words that sound alike but are spelled differently should end up closer than words spelled alike but pronounced differently, because the model is trained on sound similarity and uses phonetic rather than orthographic input.
The learned distances can be used to rerank ASR hypotheses, since the shared space gives a direct acoustic-to-phonetic match score for candidate words.
The self-labeling loop grows the training set as training proceeds (the paper observes 41% relative growth), which means the contrastive loss can keep improving without new human annotation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Because the paper's only reported evaluation is against pipeline-generated labels, I infer that the practical value of the embeddings still needs to be demonstrated on a downstream task whose ground truth does not depend on the same forced-alignment pipeline.
The same weak-supervision recipe—positive pairs from one noisy signal, hard negatives from substitution errors, and self-labeling from a k-d tree—could plausibly transfer to other bimodal embedding problems where one modality is easy to enumerate, such as text-to-image matching.
A testable extension is to train on the same data with orthographic spellings instead of phonetic ones; if homograph pairs behave as the paper expects, the phonetic-trained model should separate words like the metal 'lead' from the verb 'lead' while the orthographic model cannot.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Two things you should know before reading it. First, the paper has a genuinely useful idea: train a Siamese net on (audio segment, ARPAbet phonetic spelling) pairs using a contrastive loss, and mine hard negatives semi-supervised by embedding the phonetic lexicon and augmenting the training set with the model's own mistakes. That technique is worth taking seriously. Second, the headline number—F1 0.95 on a binary same-word/different-word task—is probably inflated by a data-splitting artifact. An example-level split means the same audio waveform can appear in both training and test, so the acoustic encoder can memorize the test inputs.

The paper does some things well. The writing is clear, the data pipeline is described in enough detail to spot the problem, and the architecture comparison (CNN vs. LSTM, dropout, margin, etc.) is honest in the sense that it reports failures too. The self-labeling mining scheme is the most original piece: sample audio segments, query the phonetic embedding tree, and add the model's newly-made mistakes to the training set. That could transfer to other embedding tasks.

Now the soft spots, in descending order of severity. The train/dev/test split: the text says the 654,224 examples were 'split, with class-label stratification,' and then reports 352,769/11,775/17,632 unique audio segments per split. Since the same audio segment is reused in many examples (one positive, many negatives), an example-level split almost guarantees overlap of identical acoustic inputs across splits. The paper never states that the sets are disjoint by audio segment, recording, or speaker. This is load-bearing: even with perfect labels, the model can score test pairs by recognizing exact waveforms, and the reported F1 says little about generalization to unseen speech. This is the first thing I'd ask the authors about. Second, the distance threshold for test labels was set based on the test-set break-even point (Section 5), which is a mild but real form of test-set leakage. Third, no code, data, baselines, or error bars; the dataset is proprietary. The label generation relies on ASR alignments, which could introduce noise, but that's a secondary issue given the split problem.

Who is this for? People working on acoustic word embeddings who want to see a practical hard-negative mining recipe. The method isn't a breakthrough, but it's a legitimate extension of [16] and [45]. As it stands, the central claim isn't credible. The fix is straightforward: re-run with a segment-disjoint split (and ideally a public benchmark), report variability, and show the test threshold selection was not informed by test labels.

My recommendation: it deserves a serious referee—not a desk reject—because the core idea is sound and the flaw is checkable and fixable. But I would not cite the F1 result until it survives a proper split.

Referee Report

4 major / 6 minor

Summary. The paper proposes a weakly supervised method for learning joint acoustic-phonetic word embeddings. It trains Siamese networks on binary same-word/different-word pairs consisting of an audio segment and a phonetic spelling, using contrastive loss with hard-negative mining, including a self-labeling technique. The authors report a best test F1 of 0.95 for a Siamese CNN (Table 1, row 9) on a proprietary dataset of 25k recordings, and claim this demonstrates a viable weakly supervised approach to learning word-sound similarity in a shared latent space.

Significance. If the reported result is reliable, the paper makes a useful empirical contribution: it shows that a binary classification surrogate, combined with hard-negative mining and self-labeling, can learn acoustic-phonetic embeddings without manual pairwise similarity labels. The paper includes a broad architecture comparison (CNNs, LSTMs, bidirectional LSTMs), several contrastive-loss variants, and a qualitative error analysis that identifies failure modes such as noisy audio, far-field speech, and accented speech. These are strengths. However, the headline F1=0.95 is a single reported number on a proprietary dataset, and two methodological concerns currently prevent me from accepting it at face value: the decision threshold appears to be selected using test-set labels, and the data split may allow the same audio segment to appear in both training and test. Because the dataset and code are not public, these issues cannot be checked by readers, so the authors must address them explicitly and re-report the key numbers.

major comments (4)

[Section 5, first paragraph] The reported F1=0.95 uses a distance threshold of 0.5, which the authors state 'approximates the observed break-even point' computed from all unique pairs in the test set (approximately 192.6 million word pairs). Selecting the decision threshold using test-set labels is an information leak and can inflate the reported score. Please fix the threshold using only the development set (or a nested validation procedure) and re-report the F1 numbers for all models in Table 1.
[Section 3, dataset split paragraph] The dataset split is described at the example level with class-label stratification, and the paper reports that the train/dev/test sets contain 352,769/11,775/17,632 unique audio segments, respectively. The paper never states that these sets are disjoint by audio segment, recording, or speaker. Because 654,224 examples are derived from a smaller number of audio segments, an example-level split can place the same waveform in both training and test. The acoustic encoder could then memorize the exact test waveforms and obtain F1=0.95 without generalizing to unseen speech. Please re-split by unique audio segment (ideally by recording or speaker) and report the test F1 under that split.
[Section 3, label-generation paragraph] The training and evaluation labels are generated automatically by force-aligning ensemble ASR hypotheses with human transcripts; positive pairs are aligned matches, and hard negatives are ASR substitution errors and synthetic phonetic-edit-distance neighbors. If the ASR or aligner systematically mislabels words, the learned embeddings are trained and evaluated against the same flawed labels. The paper should quantify label quality, for example by manually inspecting a random subset of positive and hard-negative pairs, or by reporting agreement with human similarity judgments on a held-out set. This is load-bearing because the label-generation procedure is the only supervision signal, and self-labeling further propagates these labels.
[Section 5, Table 1] The headline F1=0.95 is a single run on a single test set, with no error bars, confidence intervals, or repeated runs. Given that Table 1 shows nontrivial variation across configurations (e.g., rows 3 and 4 differ by 0.03 on test F1), the authors should report bootstrap confidence intervals or results over multiple random seeds so that readers can assess whether 0.95 is significantly better than, say, 0.94 or 0.93. This is especially important because the dataset is proprietary and not available for independent evaluation.

minor comments (6)

[Section 3, first paragraph of audio representation] The text says 'mel-spectograms'; this should be 'mel-spectrograms'.
[Section 6, first paragraph] The sentence 'The choice of training examples cannot be cannot be overstated' contains a duplicated phrase and should read 'cannot be overstated'.
[Table 1] The table header 'F1 Scores' does not clearly distinguish the two reported columns as training and test F1. Please label the columns explicitly, for example 'F1 (train)' and 'F1 (test)'.
[Figure 1] Figure 1 is not referenced in the text; please add a sentence describing what the word-duration and word-frequency distributions show and how they influenced the preprocessing choices.
[Section 3, hard-negative mining] The term 'break-even point' is used without definition, and the 'maximum phonetic-edit distance' threshold of 0.7 and the self-labeling 'maximum distance that reflects similarity' are stated but not tied to a specific tuning procedure. Please define these terms and specify how the thresholds were selected.
[References] Several references are incomplete: [7], [8], and [9] lack publication venues, and [45] is missing the page range or DOI. Please provide full citation information.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported F1 is an empirical benchmark on held-out test pairs, and no fitted quantity is relabeled as a prediction.

full rationale

The paper is an empirical study rather than a derivation, and the central claim is an F1 score on a held-out test set. Training uses a contrastive loss on weakly labeled acoustic-phonetic pairs, while evaluation computes F1 at a fixed distance threshold on all unique test pairs; the training objective and evaluation metric are related but not identical, and no fitted parameter is renamed as a prediction. The self-labeling technique augments training data with hard negatives discovered by the model, but each candidate pair's true label is looked up from ground-truth transcripts and forced alignments, so the model does not label its own evaluation data. The 0.5 threshold is described as approximating the observed break-even point for margin m = 1, which is a hyperparameter choice, not a quantity fitted to the test F1. Citations to related work are background and methodological references, not load-bearing self-citations or imported uniqueness theorems. The potential train/test overlap of audio segments is a data-leakage and generalization concern, not a circularity of the argument. No equation or fitted parameter reduces by construction to its own input, so the appropriate finding is no significant circularity.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several manually chosen thresholds and domain assumptions about label quality. There are no invented entities. The listed free parameters are chosen by hand or tuned on the data; none is derived from first principles.

free parameters (6)

margin m in contrastive loss = 1
Controls when dissimilar pairs contribute to loss; chosen by hand and used across experiments.
distance threshold for binary labels = 0.5
Approximates observed break-even point; used at test time to convert distances to labels. May have been selected with knowledge of test set.
maximum phonetic-edit distance for synthetic negatives = 0.7
Threshold for selecting ASR substitution pairs to synthesize new negative examples; chosen manually as satisfactory.
minimum word duration = 0.2 seconds
Words shorter than this were removed because manual inspection deemed them mostly mislabeled.
zero-padding window = 2 seconds
Audio inputs padded with silence to a fixed 2-second window; chosen to fix the model input size, and it affects all inputs.
embedding dimension (best model) = 512
Model hyperparameter tuned manually; larger embeddings may affect discrimination ability.

assumptions (4)

domain assumption ASR hypotheses and human transcripts align accurately enough to produce correct positive and negative word-pair labels.
Section 3: labels are derived from forced alignment between ASR ensemble and human transcripts; errors in alignment would corrupt supervision.
domain assumption LibriSpeech lexicon coverage is sufficient and OOV words can be excluded without biasing the dataset.
Section 3: OOV words are excluded because they lack phonetic labels; this shapes the vocabulary and may remove hard cases.
domain assumption Zero-padding to 2 seconds preserves word identity for all words in the dataset.
Section 3: audio is padded with silence to fit a 2-second window; if any word exceeds 2 seconds, part of it is lost, which is not discussed.
ad hoc to paper Manual inspection of word durations is a valid basis for discarding data.
Section 3: words shorter than 0.2 seconds are removed as mostly mislabeled based on subjective manual inspection, with no quantitative validation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Learning Joint Acoustic-Phonetic Word Embeddings." pith.science (2026). https://pith.science/paper/UVP52RCN

@misc{pith2026190800493,
  author       = {Pith},
  title        = {Pith review of: Learning Joint Acoustic-Phonetic Word Embeddings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UVP52RCN}},
  note         = {Machine review of arXiv:1908.00493}
}

abstract

Most speech recognition tasks pertain to mapping words across two modalities: acoustic and orthographic. In this work, we suggest learning encoders that map variable-length, acoustic or phonetic, sequences that represent words into fixed-dimensional vectors in a shared latent space; such that the distance between two word vectors represents how closely the two words sound. Instead of directly learning the distances between word vectors, we employ weak supervision and model a binary classification task to predict whether two inputs, one of each modality, represent the same word given a distance threshold. We explore various deep-learning models, bimodal contrastive losses, and techniques for mining hard negative examples such as the semi-supervised technique of self-labeling. Our best model achieves an $F_1$ score of 0.95 for the binary classification task.

Figures

Figures reproduced from arXiv: 1908.00493 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 44 canonical work pages

[16]

Multi-View Recurrent Neural Acoustic Word Embeddings

Wanjia He, Weiran Wang, and Karen Livescu. “Multi-View Recurrent Neural Acoustic Word Embeddings”. In: International Conference on Learning Representations (2017)

work page 2017
[45]

Phonetics embedding learning with side information

Gabriel Synnaeve, Thomas Schatz, and Emmanuel Dupoux. “Phonetics embedding learning with side information”. In: 2014 IEEE Spoken Language Technology Workshop (SLT). IEEE. 2014, pp. 106–111

work page 2014
[1]

Introducing DeepText: Facebook’s Text Understanding Engine

Ahmad Abdulkader, A Lakshmiratan, and J Zhang. “Introducing DeepText: Facebook’s Text Understanding Engine”. In: Facebook Code (2016)

work page 2016
[2]

Word Embeddings for Speech Recognition

Samy Bengio and Georg Heigold. “Word Embeddings for Speech Recognition”. In:Fifteenth Annual Conference of the International Speech Communication Association. 2014

work page 2014
[3]

Natural Language Processing with Python

Steven Bird, Ewan Klein, and Edward Loper. Natural Language Processing with Python. 1st. O’Reilly Media, Inc., 2009. ISBN : 9780596516499

work page 2009
[4]

Signature Veriﬁcation Using a

Jane Bromley et al. “Signature Veriﬁcation Using a "Siamese" Time Delay Neural Network”. In: Advances in neural information processing systems. 1994, pp. 737–744

work page 1994
[5]

Query-by-Example Keyword Spotting Using Long Short-Term Memory Networks

Guoguo Chen, Carolina Parada, and Tara N Sainath. “Query-by-Example Keyword Spotting Using Long Short-Term Memory Networks”. In: 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 2015, pp. 5236–5240

work page 2015
[6]

François Chollet et al. Keras. https://keras.io. 2015

work page 2015

Show all 47 references

[7]

Learning Word Embeddings from Speech

Yu-An Chung and James Glass. “Learning Word Embeddings from Speech”. In: (2017)

2017
[8]

Speech2vec: A Sequence-to-Sequence Framework for Learn- ing Word Embeddings from Speech

Yu-An Chung and James Glass. “Speech2vec: A Sequence-to-Sequence Framework for Learn- ing Word Embeddings from Speech”. In:INTERSPEECH. 2018

2018
[9]

Audio Word2Vec: Unsupervised Learning of Audio Segment Representa- tions Using Sequence-to-Sequence Autoencoder

Yu-An Chung et al. “Audio Word2Vec: Unsupervised Learning of Audio Segment Representa- tions Using Sequence-to-Sequence Autoencoder”. In: (2016)

2016
[10]

Unsupervised Cross-Modal Alignment of Speech and Text Embedding Spaces

Yu-An Chung et al. “Unsupervised Cross-Modal Alignment of Speech and Text Embedding Spaces”. In: Advances in Neural Information Processing Systems 31. Ed. by S. Bengio et al. Curran Associates, Inc., 2018, pp. 7354–7364. URL: http://papers.nips.cc/paper/ 7965-unsupervised-cros...

2018
[11]

FFmpeg. 2016. URL: http://http://www.ffmpeg.org/ (visited on 12/16/2016)

2016
[12]

Understanding the difﬁculty of training deep feedforward neural networks

Xavier Glorot and Yoshua Bengio. “Understanding the difﬁculty of training deep feedforward neural networks”. In: Proceedings of the thirteenth international conference on artiﬁcial intelligence and statistics. 2010, pp. 249–256

2010
[13]

Dimensionality Reduction by Learning an Invariant Mapping

Raia Hadsell, Sumit Chopra, and Yann LeCun. “Dimensionality Reduction by Learning an Invariant Mapping”. In: 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’06). V ol. 2. IEEE. 2006, pp. 1735–1742

2006
[14]

Learning Word-Like Units from Joint Audio-Visual Analysis

David Harwath and James R Glass. “Learning Word-Like Units from Joint Audio-Visual Analysis”. In: (2017)

2017
[15]

Delving deep into rectiﬁers: Surpassing human-level performance on imagenet classiﬁcation

Kaiming He et al. “Delving deep into rectiﬁers: Surpassing human-level performance on imagenet classiﬁcation”. In: Proceedings of the IEEE international conference on computer vision. 2015, pp. 1026–1034

2015
[17]

Matplotlib: A 2D graphics environment

John D Hunter. “Matplotlib: A 2D graphics environment”. In: Computing in science & engineering 9.3 (2007), p. 90

2007
[18]

{SciPy}: Open source scientiﬁc tools for {Python}

Eric Jones, Travis Oliphant, and Pearu Peterson. “{SciPy}: Open source scientiﬁc tools for {Python}”. In: (2014)

2014
[19]

Kaldi ASR. 2016. URL: http://kaldi-asr.org/

2016
[20]

A Segmental Framework for Fully- Unsupervised Large-V ocabulary Speech Recognition

Herman Kamper, Aren Jansen, and Sharon Goldwater. “A Segmental Framework for Fully- Unsupervised Large-V ocabulary Speech Recognition”. In:Computer Speech & Language 46 (2017), pp. 154–174. 6

2017
[21]

Deep Convolutional Acoustic Word Embeddings Using Word-Pair Side Information

Herman Kamper, Weiran Wang, and Karen Livescu. “Deep Convolutional Acoustic Word Embeddings Using Word-Pair Side Information”. In:Acoustics, Speech and Signal Processing (ICASSP), 2016 IEEE International Conference on. IEEE. 2016, pp. 4950–4954

2016
[22]

Adam: A method for stochastic optimization

Diederik P Kingma and Jimmy Ba. “Adam: A method for stochastic optimization”. In: (2014)

2014
[23]

Jupyter Notebooks – a publishing format for reproducible computa- tional workﬂows

Thomas Kluyver et al. “Jupyter Notebooks – a publishing format for reproducible computa- tional workﬂows”. In: Positioning and Power in Academic Publishing: Players, Agents and Agendas. Ed. by F. Loizides and B. Schmidt. IOS Press. 2016, pp. 87–90

2016
[24]

Segmental Acoustic Indexing for Zero Resource Keyword Search

Keith Levin, Aren Jansen, and Benjamin Van Durme. “Segmental Acoustic Indexing for Zero Resource Keyword Search”. In: 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 2015, pp. 5828–5832

2015
[25]

Fixed-Dimensional Acoustic Embeddings of Variable-Length Segments in Low-Resource Settings

Keith Levin et al. “Fixed-Dimensional Acoustic Embeddings of Variable-Length Segments in Low-Resource Settings”. In: Automatic Speech Recognition and Understanding (ASRU), 2013 IEEE Workshop on. IEEE. 2013, pp. 410–415

2013
[26]

A Survey of Multi-View Representation Learning

Yingming Li, Ming Yang, and Zhongfei Mark Zhang. “A Survey of Multi-View Representation Learning”. In: IEEE Transactions on Knowledge and Data Engineering (2018)

2018
[27]

Learning Acoustic Word Embeddings with Phonetically Associated Triplet Network

Hyungjun Lim et al. “Learning Acoustic Word Embeddings with Phonetically Associated Triplet Network”. In: (2018)

2018
[28]

ASR Hypothesis Reranking Using Prior- Informed Restricted Boltzmann Machine

Yukun Ma, Erik Cambria, and Benjamin Bigot. “ASR Hypothesis Reranking Using Prior- Informed Restricted Boltzmann Machine”. In: International Conference on Computational Linguistics and Intelligent Text Processing. Springer. 2017, pp. 503–514

2017
[29]

Visualizing data using t-SNE

Laurens van der Maaten and Geoffrey Hinton. “Visualizing data using t-SNE”. In:Journal of machine learning research 9.Nov (2008), pp. 2579–2605

2008
[30]

librosa/librosa: 0.6.3

Brian McFee et al. librosa/librosa: 0.6.3. Feb. 2019. DOI: 10.5281/zenodo.2564164. URL: https://doi.org/10.5281/zenodo.2564164

2019 doi
[31]

Data structures for statistical computing in python

Wes McKinney et al. “Data structures for statistical computing in python”. In:Proceedings of the 9th Python in Science Conference. V ol. 445. Austin, TX. 2010, pp. 51–56

2010
[32]

Distributed Representations of Words and Phrases and Their Composi- tionality

Tomas Mikolov et al. “Distributed Representations of Words and Phrases and Their Composi- tionality”. In: Advances in neural information processing systems. 2013, pp. 3111–3119

2013
[33]

Rectiﬁed linear units improve restricted boltzmann machines

Vinod Nair and Geoffrey E Hinton. “Rectiﬁed linear units improve restricted boltzmann machines”. In: Proceedings of the 27th international conference on machine learning (ICML- 10). 2010, pp. 807–814

2010
[34]

Librispeech: An ASR Corpus Based on Public Domain Audio Books

V . Panayotov et al. “Librispeech: An ASR Corpus Based on Public Domain Audio Books”. In: 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Apr. 2015, pp. 5206–5210. DOI: 10.1109/ICASSP.2015.7178964

2015
[35]

Automatic differentiation in PyTorch

Adam Paszke et al. “Automatic differentiation in PyTorch”. In: (2017)

2017
[36]

Scikit-learn: Machine Learning in Python

F. Pedregosa et al. “Scikit-learn: Machine Learning in Python”. In: Journal of Machine Learning Research 12 (2011), pp. 2825–2830

2011
[37]

GloVe: Global Vectors for Word Representation

Jeffrey Pennington, Richard Socher, and Christopher Manning. “GloVe: Global Vectors for Word Representation”. In:Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP). 2014, pp. 1532–1543

2014
[38]

Facenet: A Uniﬁed Embedding for Face Recognition and Clustering

Florian Schroff, Dmitry Kalenichenko, and James Philbin. “Facenet: A Uniﬁed Embedding for Face Recognition and Clustering”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2015, pp. 815–823

2015
[39]

Probability of error of some adaptive pattern-recognition machines

H Scudder. “Probability of error of some adaptive pattern-recognition machines”. In:IEEE Transactions on Information Theory 11.3 (1965), pp. 363–371

1965
[40]

Discriminative Acoustic Word Embeddings: Recurrent Neural Network-Based Approaches

S. Settle and K. Livescu. “Discriminative Acoustic Word Embeddings: Recurrent Neural Network-Based Approaches”. In: 2016 IEEE Spoken Language Technology Workshop (SLT). Dec. 2016, pp. 503–510

2016
[41]

Query-by-Example Search with Discriminative Neural Acoustic Word Embeddings

Shane Settle et al. “Query-by-Example Search with Discriminative Neural Acoustic Word Embeddings”. In: (2017)

2017
[42]

SoX - Sound eXchange . 2015. URL: http : / / sox . sourceforge . net/ (visited on 12/16/2016). 7

2015
[43]

Dropout: a simple way to prevent neural networks from overﬁtting

Nitish Srivastava et al. “Dropout: a simple way to prevent neural networks from overﬁtting”. In: The Journal of Machine Learning Research 15.1 (2014), pp. 1929–1958

2014
[44]

Cepstral Mean and Variance Normalization in the Model Domain

Ole Morten Strand and Andreas Egeberg. “Cepstral Mean and Variance Normalization in the Model Domain”. In: COST278 and ISCA Tutorial and Research Workshop (ITRW) on Robustness Issues in Conversational Interaction. 2004

2004
[46]

Learning Deep Embeddings with Histogram Loss

Evgeniya Ustinova and Victor Lempitsky. “Learning Deep Embeddings with Histogram Loss”. In: Advances in Neural Information Processing Systems 29. Ed. by D. D. Lee et al. Curran Associates, Inc., 2016, pp. 4170–4178

2016
[47]

Sampling Matters in Deep Embedding Learning

Chao-Yuan Wu et al. “Sampling Matters in Deep Embedding Learning”. In:Proceedings of the IEEE International Conference on Computer Vision. 2017, pp. 2840–2848. 8

2017

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [16]

Multi-View Recurrent Neural Acoustic Word Embeddings

Wanjia He, Weiran Wang, and Karen Livescu. “Multi-View Recurrent Neural Acoustic Word Embeddings”. In: International Conference on Learning Representations (2017)

work page 2017

[2] [45]

Phonetics embedding learning with side information

Gabriel Synnaeve, Thomas Schatz, and Emmanuel Dupoux. “Phonetics embedding learning with side information”. In: 2014 IEEE Spoken Language Technology Workshop (SLT). IEEE. 2014, pp. 106–111

work page 2014

[3] [1]

Introducing DeepText: Facebook’s Text Understanding Engine

Ahmad Abdulkader, A Lakshmiratan, and J Zhang. “Introducing DeepText: Facebook’s Text Understanding Engine”. In: Facebook Code (2016)

work page 2016

[4] [2]

Word Embeddings for Speech Recognition

Samy Bengio and Georg Heigold. “Word Embeddings for Speech Recognition”. In:Fifteenth Annual Conference of the International Speech Communication Association. 2014

work page 2014

[5] [3]

Natural Language Processing with Python

Steven Bird, Ewan Klein, and Edward Loper. Natural Language Processing with Python. 1st. O’Reilly Media, Inc., 2009. ISBN : 9780596516499

work page 2009

[6] [4]

Signature Veriﬁcation Using a

Jane Bromley et al. “Signature Veriﬁcation Using a "Siamese" Time Delay Neural Network”. In: Advances in neural information processing systems. 1994, pp. 737–744

work page 1994

[7] [5]

Query-by-Example Keyword Spotting Using Long Short-Term Memory Networks

Guoguo Chen, Carolina Parada, and Tara N Sainath. “Query-by-Example Keyword Spotting Using Long Short-Term Memory Networks”. In: 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 2015, pp. 5236–5240

work page 2015

[8] [6]

François Chollet et al. Keras. https://keras.io. 2015

work page 2015