We introduce Hibiki, a decoder-only model for simultaneous speech translation. Hibiki leverages a multistream language model to synchronously process source and target speech, and jointly produces text and audio tokens to perform speech-to-text and speech-to-speech translation. We furthermore address the fundamental challenge of simultaneous interpretation, which unlike its consecutive counterpart --where one waits for the end of the source utterance to start translating-- adapts its flow to accumulate just enough context to produce a correct translation in real-time, chunk by chunk. To do so, we introduce a weakly-supervised method that leverages the perplexity of an off-the-shelf text translation system to identify optimal delays on a per-word basis and create aligned synthetic data. After supervised training, Hibiki performs adaptive, simultaneous speech translation with vanilla temperature sampling. On a French-English simultaneous speech translation task, Hibiki demonstrates state-of-the-art performance in translation quality, speaker fidelity and naturalness. Moreover, the simplicity of its inference process makes it compatible with batched translation and even real-time on-device deployment. We provide examples on *huggingface.co/spaces/kyutai/hibiki-samples* as well as models and inference code at *github.com/kyutai-labs/hibiki*.

We propose a novel preference alignment framework for improving spoken dialogue models on real-time conversations from user interactions. Current preference learning methods primarily focus on text-based language models, and are not directly suited to the complexities of real-time speech interactions, with richer dynamics (e.g. interruption, interjection) and no explicit segmentation between speaker turns.We create a large-scale dataset of more than 150,000 preference pairs from raw multi-turn speech conversations, annotated with AI feedback, to cover preferences over both linguistic content and temporal context variations. We leverage offline alignment methods to finetune a full-duplex autoregressive speech-to-speech model. Extensive experiments demonstrate that feedback on generic conversations can be consistently effective in improving spoken dialogue models to produce more factual, safer and more contextually aligned interactions. We deploy the finetuned model and conduct holistic human evaluations to assess the impact beyond single-turn conversations. Our findings shed light on the importance of a well-calibrated balance among various dynamics, crucial for natural real-time speech dialogue systems.

Matthieu Futeral, Andrea Agostinelli, Marco Tagliasacchi, Neil Zeghidour, Eugene Kharitonov. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

We propose MusicRL, the first music generation system finetuned from human feedback. Appreciation of text-to-music models is particularly subjective since the concept of musicality as well as the specific intention behind a caption are user-dependent (e.g. a caption such as “upbeat workout music” can map to a retro guitar solo or a technopop beat). Not only this makes supervised training of such models challenging, but it also calls for integrating continuous human feedback in their post-deployment finetuning. MusicRL is a pretrained autoregressive MusicLM model of discrete audio tokens finetuned with reinforcement learning to maximize sequence-level rewards. We design reward functions related specifically to text-adherence and audio quality with the help from selected raters, and use those to finetune MusicLM into MusicRL-R. We deploy MusicLM to users and collect a substantial dataset comprising 300,000 pairwise preferences. Using Reinforcement Learning from Human Feedback (RLHF), we train MusicRL-U, the first text-to-music model that incorporates human feedback at scale. Human evaluations show that both MusicRL-R and MusicRL-U are preferred to the baseline. Ultimately, MusicRL-RU combines the two approaches and results in the best model according to human raters. Ablation studies shed light on the musical attributes influencing human preferences, indicating that text adherence and quality only account for a part of it. This underscores the prevalence of subjectivity in musical appreciation and calls for further involvement of human listeners in the finetuning of music generation models. Samples can be found at google-research.github.io/seanet/musiclm/rlhf/.

We introduce LMCodec, a causal neural speech codec that provides high quality audio at very low bitrates. The backbone of the system is a causal convolutional codec that encodes audio into a hierarchy of coarse-to-fine tokens using residual vector quantization. LMCodec trains a Transformer language model to predict the fine tokens from the coarse ones in a generative fashion, allowing for the transmission of fewer codes. A second Transformer predicts the uncertainty of the next codes given the past transmitted codes, and is used to perform conditional entropy coding. A MUSHRA subjective test was conducted and shows that the quality is comparable to reference codecs at higher bitrates. Example audio is available at https://mjenrungrot.github.io/chrome-media-audio-papers/publications/lmcodec.

We present a method to separate speech signals from noisy environments in the embedding space of a neural audio codec. We introduce a new training procedure that allows our model to produce structured encodings of audio waveforms given by embedding vectors, where one part of the embedding vector represents the speech signal, and the rest represent the environment. We achieve this by partitioning the embeddings of different input waveforms and training the model to faithfully reconstruct audio from mixed partitions, thereby ensuring each partition encodes a separate audio attribute. As use cases, we demonstrate the separation of speech from background noise or from reverberation characteristics. Our method also allows for targeted adjustments of the audio output characteristics.

We developed dysarthric speech intelligibility classifiers on 551,176 disordered speech samples contributed by a diverse set of 468 speakers, with a range of self-reported speaking disorders and rated for their overall intelligibility on a five-point scale. We trained three models following different deep learning approaches and evaluated them on ~ 94K utterances from 100 speakers. We further found the models to generalize well (without further training) on the TORGO database[1] (100% accuracy), UASpeech[2] (0.93 correlation), ALS-TDI PMP[3] (0.81 AUC) datasets as well as on a dataset of realistic unprompted speech we gathered (106 dysarthric and 76 control speakers, ~ 2300 samples). We share our model1to advance research in this domain.

This paper presents a Graph-based Simultaneous Localization And Mapping (GraphSLAM) approach for a robotic system relying on the reflections of ultrasonic guided waves to enable long-range inspection tasks on plate-based metal structures. A measurement model that can leverage multi-order acoustic echoes is introduced for accurate localization, and beamforming is used for mapping the boundaries of individual metal panels. These two elements are subsequently integrated within a nonlinear least squares optimizer to solve the full offline SLAM problem. We experimentally evaluate the potential of this approach in a laboratory environment. We observe the improved localization accuracy of the multi-order echo model compared to a second model, from previous works, that relies solely on first-order echoes. We also show that the proposed approach can yield accurate SLAM results, hence showcasing the standalone capability of ultrasonic-based GraphSLAM for envisioned long-range inspection applications.

Abstract We introduce SPEAR-TTS, a multi-speaker text-to-speech (TTS) system that can be trained with minimal supervision. By combining two types of discrete speech representations, we cast TTS as a composition of two sequence-to-sequence tasks: from text to high-level semantic tokens (akin to “reading”) and from semantic tokens to low-level acoustic tokens (“speaking”). Decoupling these two tasks enables training of the “speaking” module using abundant audio-only data, and unlocks the highly efficient combination of pretraining and backtranslation to reduce the need for parallel data when training the “reading” component. To control the speaker identity, we adopt example prompting, which allows SPEAR-TTS to generalize to unseen speakers using only a short sample of 3 seconds, without any explicit speaker representation or speaker labels. Our experiments demonstrate that SPEAR-TTS achieves a character error rate that is competitive with state-of-the-art methods using only 15 minutes of parallel data, while matching ground-truth speech in naturalness and acoustic quality.

We introduce AudioLM, a framework for high-quality audio generation with long-term consistency. AudioLM maps the input audio to a sequence of discrete tokens and casts audio generation as a language modeling task in this representation space. We show how existing audio tokenizers provide different trade-offs between reconstruction quality and long-term structure, and we propose a hybrid tokenization scheme to achieve both objectives. Namely, we leverage the discretized activations of a masked language model pre-trained on audio to capture long-term structure and the discrete codes produced by a neural audio codec to achieve high-quality synthesis. By training on large corpora of raw audio waveforms, AudioLM learns to generate natural and coherent continuations given short prompts. When trained on speech, and without any transcript or annotation, AudioLM generates syntactically and semantically plausible speech continuations while also maintaining speaker identity and prosody for unseen speakers. Furthermore, we demonstrate how our approach extends beyond speech by generating coherent piano music continuations, despite being trained without any symbolic representation of music.

Real-world data is high-dimensional: a book, image, or musical performance can easily contain hundreds of thousands of elements even after compression. However, the most commonly used autoregressive models, Transformers, are prohibitively expensive to scale to the number of inputs and layers needed to capture this long-range structure. We develop Perceiver AR, an autoregressive, modality-agnostic architecture which uses cross-attention to map long-range inputs to a small number of latents while also maintaining end-to-end causal masking. Perceiver AR can directly attend to over a hundred thousand tokens, enabling practical long-context density estimation without the need for hand-crafted sparsity patterns or memory mechanisms. When trained on images or music, Perceiver AR generates outputs with clear long-term coherence and structure. Our architecture also obtains state-of-the-art likelihood on long-sequence benchmarks, including 64x64 ImageNet images and PG-19 books.

The ultrasonic mapping of plate-based facilities is an essential step towards the robotic inspection of large metal structures such as storage tanks or ship hulls. This work proposes a novel framework that exploits ultrasonic echoes to recover grid-based and feature-based spatial representations jointly. We aim to improve on a previous mapping method [1] subject to errors due to interference, and which provides plate geometry estimates without uncertainty assessment. The grid can represent, all along the mapping process, both areas identified as inside or outside the current plate and areas whose state is still unknown, making it is suitable e.g. for detecting a change of plate, or for use in a later active-sensing strategy. We also leverage the resulting spatial information to filter out candidate plate edges that are no longer relevant, mitigating the detrimental effect of interference. We test the approach in simulation, with acoustic data acquired manually and with a real robot. Results show that it is effective for building combined map representations and robust to echo misdetection, contrary to a more standard mapping approach.

Deep audio classification, traditionally cast as training a deep neural network on top of mel-filterbanks in a supervised fashion, has recently benefited from two independent lines of work.The first one explores "learnable frontends", i.e., neural modules that produce a learnable time-frequency representation, to overcome limitations of fixed features.The second one uses self-supervised learning to leverage unprecedented scales of pre-training data.In this work, we study the feasibility of combining both approaches, i.e., pre-training learnable frontend jointly with the main architecture for downstream classification.First, we show that pretraining two previously proposed frontends (SincNet and LEAF) on Audioset drastically improves linear-probe performance over fixed mel-filterbanks, suggesting that learnable time-frequency representations can benefit self-supervised pre-training even more than supervised training.Surprisingly, randomly initialized learnable filterbanks outperform mel-scaled initialization in the self-supervised setting, a counter-intuitive result that questions the appropriateness of strong priors when designing learnable filters.Through exploratory analysis of the learned frontend components, we uncover crucial differences in properties of these frontends when used in a supervised and self-supervised setting, especially the affinity of self-supervised filters to diverge significantly from the mel-scale to model a broader range of frequencies.

We presentSoundStream, a novel neural audio codec that can efficiently compress speech, music and general audio at bitrates normally targeted by speech-tailored codecs.SoundStreamrelies on a model architecture composed by a fully convolutional encoder/decoder network and a residual vector quantizer, which are trained jointly end-to-end. Training leverages recent advances in text-to-speech and speech enhancement, which combine adversarial and reconstruction losses to allow the generation of high-quality audio content from quantized embeddings. By training with structured dropout applied to quantizer layers, a single model can operate across variable bitrates from 3 kbps to 18 kbps, with a negligible quality loss when compared with models trained at fixed bitrates. In addition, the model is amenable to a low latency implementation, which supports streamable inference and runs in real time on a smartphone CPU. In subjective evaluations using audio at 24 kHz sampling rate,SoundStreamat 3 kbps outperforms Opus at 12 kbps and approaches EVS at 9.6 kbps. Moreover, we are able to perform joint compression and enhancement either at the encoder or at the decoder side with no additional latency, which we demonstrate through background noise suppression for speech.

We introduce DIVE, an end-to-end speaker diarization sys-tem. DIVE presents the diarization task as an iterative pro-cess: it repeatedly builds a representation for each speaker before predicting their voice activity conditioned on the ex-tracted representations. This strategy intrinsically resolves the speaker ordering ambiguity without requiring the classi-cal permutation invariant training loss. In contrast with prior work, our model does not rely on pretrained speaker represen-tations and jointly optimizes all parameters of the system with a multi-speaker voice activity loss. DIVE does not require the training speaker identities and allows efficient window-based training. Importantly, our loss explicitly excludes unreliable speaker turn boundaries from training, which is adapted to the standard collar-based Diarization Error Rate (DER) eval-uation. Overall, these contributions yield a system redefining the state-of-the-art on the CALLHOME benchmark, with 6.7% DER compared to 7.8% for the best alternative.

We propose CHARM, a method for training a single neural network across inconsistent input channels. Our work is motivated by Electroencephalography (EEG), where data collection protocols from different headsets result in varying channel ordering and number, which limits the feasibility of transferring trained systems across datasets. Our approach builds upon attention mechanisms to estimate a latent reordering matrix from each input signal and map input channels to a canonical order. CHARM is differentiable and can be composed further with architectures expecting a consistent channel ordering to build end-to-end trainable classifiers. We perform experiments on four EEG classification datasets and demonstrate the efficacy of CHARM via simulated shuffling and masking of input channels. Moreover, our method improves the transfer of pre-trained representations between datasets collected with different protocols.

We introduce COLA, a self-supervised pre-training approach for learning a general-purpose representation of audio. Our approach is based on contrastive learning: it learns a representation which assigns high similarity to audio segments extracted from the same recording while assigning lower similarity to segments from different recordings. We build on top of recent advances in contrastive learning for computer vision and reinforcement learning to design a lightweight, easy-to-implement self-supervised model of audio. We pre-train embeddings on the large-scale Audioset database and transfer these representations to 9 diverse classification tasks, including speech, music, animal sounds, and acoustic scenes. We show that despite its simplicity, our method significantly outperforms previous self-supervised systems. We furthermore conduct ablation studies to identify key design choices and release a library1to pre-train and fine-tune COLA models.

Self-supervised pre-training using so-called “pretext” tasks has recently shown impressive performance across a wide range of modalities. In this work, we advance self-supervised learning from permutations, by pre-training a model to reorder shuffled parts of the spectrogram of an audio signal, to improve downstream classification performance. We make two main contributions. First, we overcome the main challenges of integrating permutation inversions into an end-to-end training scheme, using recent advances in differentiable ranking. This was heretofore sidestepped by casting the reordering task as classification, fundamentally reducing the space of permutations that can be exploited. Our experiments validate that learning from all possible permutations improves the quality of the pre-trained representations over using a limited, fixed set. Second, we show that inverting permutations is a meaningful pretext task for learning audio representations in an unsupervised fashion. In particular, we improve instrument classification and pitch estimation of musical notes by reordering spectrogram patches in the time-frequency space.

We introduce Wavesplit, an end-to-end source separation system. From a single mixture, the model infers a representation for each source and then estimates each source signal given the inferred representations. The model is trained to jointly perform both tasks from the raw waveform. Wavesplit infers a set of source representations via clustering, which addresses the fundamental permutation problem of separation. For speech separation, our sequence-wide speaker representations provide a more robust separation of long, challenging recordings compared to prior work. Wavesplit redefines the state-of-the-art on clean mixtures of 2 or 3 speakers (WSJ0-2/3mix), as well as in noisy and reverberated settings (WHAM/WHAMR). We also set a new benchmark on the recent LibriMix dataset. Finally, we show that Wavesplit is also applicable to other domains, by separating fetal and maternal heart rates from a single abdominal electrocardiogram.

Transcribed datasets typically contain speaker identity for each instance in the data. We investigate two ways to incorporate this information during training: Multi-Task Learning and Adversarial Learning. In multi-task learning, the goal is speaker prediction; we expect a performance improvement with this joint training if the two tasks of speech recognition and speaker recognition share a common set of underlying features. In contrast, adversarial learning is a means to learn representations invariant to the speaker. We then expect better performance if this learnt invariance helps generalizing to new speakers. While the two approaches seem natural in the context of speech recognition, they are incompatible because they correspond to opposite gradients back-propagated to the model. In order to better understand the effect of these approaches in terms of error rates, we compare both strategies in controlled settings. Moreover, we explore the use of additional un-transcribed data in a semi-supervised, adversarial learning manner to improve error rates. Our results show that deep models trained on big datasets already develop invariant representations to speakers without any auxiliary loss. When considering adversarial learning and multi-task learning, the impact on the acoustic model seems minor. However, models trained in a semi-supervised manner can improve error-rates.

Speech classifiers of paralinguistic traits traditionally learn from diverse hand-crafted low-level features, by selecting the relevant information for the task at hand. We explore an alternative to this selection, by learning jointly the classifier, and the feature extraction. Recent work on speech recognition has shown improved performance over speech features by learning from the waveform. We extend this approach to paralinguistic classification and propose a neural network that can learn a filterbank, a normalization factor and a compression power from the raw speech, jointly with the rest of the architecture. We apply this model to dysarthria detection from sentence-level audio recordings. Starting from a strong attention-based baseline on which mel-filterbanks outperform standard low-level descriptors, we show that learning the filters or the normalization and compression improves over fixed features by 10% absolute accuracy. We also observe a gain over OpenSmile features by learning jointly the feature extraction, the normalization, and the compression factor with the architecture. This constitutes a first attempt at learning jointly all these operations from raw audio for a speech classification task.

We train a bank of complex filters that operates on the raw waveform and is fed into a convolutional neural network for end-to-end phone recognition. These time-domain filterbanks (TD-filterbanks) are initialized as an approximation of mel-filterbanks, and then fine-tuned jointly with the remaining convolutional architecture. We perform phone recognition experiments on TIMIT and show that for several architectures, models trained on TD- filterbanks consistently outperform their counterparts trained on comparable mel-filterbanks. We get our best performance by learning all front-end steps, from pre-emphasis up to averaging. Finally, we observe that the filters at convergence have an asymmetric impulse response, and that some of them remain almost analytic.

Recent studies have investigated siamese network architectures for learning invariant speech representations using same-different side information at the word level. Here we investigate systematically an often ignored component of siamese networks: the sampling procedure (how pairs of same vs. different tokens are selected). We show that sampling strategies taking into account Zipf's Law, the distribution of speakers and the proportions of same and different pairs of words significantly impact the performance of the network. In particular, we show that word frequency compression improves learning across a large range of variations in number of training pairs. This effect does not apply to the same extent to the fully unsupervised setting, where the pairs of same-different words are obtained by spoken term discovery. We apply these results to pairs of words discovered using an unsupervised algorithm and show an improvement on state-of-the-art in unsupervised representation learning using siamese networks.

State-of-the-art speech recognition systems rely on fixed, hand-crafted features such as mel-filterbanks to preprocess the waveform before the training pipeline. In this paper, we study end-to-end systems trained directly from the raw waveform, building on two alternatives for trainable replacements of mel-filterbanks that use a convolutional architecture. The first one is inspired by gammatone filterbanks (Hoshen et al., 2015; Sainath et al, 2015), and the second one by the scattering transform (Zeghidour et al., 2017). We propose two modifications to these architectures and systematically compare them to mel-filterbanks, on the Wall Street Journal dataset. The first modification is the addition of an instance normalization layer, which greatly improves on the gammatone-based trainable filterbanks and speeds up the training of the scattering-based filterbanks. The second one relates to the low-pass filter used in these approaches. These modifications consistently improve performances for both approaches, and remove the need for a careful initialization in scattering-based trainable filterbanks. In particular, we show a consistent improvement in word error rate of the trainable filterbanks relatively to comparable mel-filterbanks. It is the first time end-to-end models trained from the raw signal significantly outperform mel-filterbanks on a large vocabulary task under clean recording conditions.

Recent progress in deep learning for audio synthesis opens the way to models that directly produce the waveform, shifting away from the traditional paradigm of relying on vocoders or MIDI synthesizers for speech or music generation. Despite their successes, current state-of-the-art neural audio synthesizers such as WaveNet and SampleRNN suffer from prohibitive training and inference times because they are based on autoregressive models that generate audio samples one at a time at a rate of 16kHz. In this work, we study the more computationally efficient alternative of generating the waveform frame-by-frame with large strides. We present a lightweight neural audio synthesizer for the original task of generating musical notes given desired instrument, pitch and velocity. Our model is trained end-to-end to generate notes from nearly 1000 instruments with a single decoder, thanks to a new loss function that minimizes the distances between the log spectrograms of the generated and target waveforms. On the generalization task of synthesizing notes for pairs of pitch and instrument not seen during training, SING produces audio with significantly improved perceptual quality compared to a state-of-the-art autoencoder based on WaveNet as measured by a Mean Opinion Score (MOS), and is about 32 times faster for training and 2, 500 times faster for inference.

Recent works have explored deep architectures for learning multimodal speech representation (e.g. audio and images, articulation and audio) in a supervised way. Here we investigate the role of combining different speech modalities, i.e. audio and visual information representing the lips movements, in a weakly supervised way using Siamese networks and lexical same-different side information. In particular, we ask whether one modality can benefit from the other to provide a richer representation for phone recognition in a weakly supervised setting. We introduce mono-task and multi-task methods for merging speech and visual modalities for phone recognition. The mono-task learning consists in applying a Siamese network on the concatenation of the two modalities, while the multi-task learning receives several different combinations of modalities at train time. We show that multi-task learning enhances discriminability for visual and multimodal inputs while minimally impacting auditory inputs. Furthermore, we present a qualitative analysis of the obtained phone embeddings, and show that cross-modal visual input can improve the discriminability of phonological features which are visually discernable (rounding, open/close, labial place of articulation), resulting in representations that are closer to abstract linguistic features than those based on audio only.

This paper introduces a new encoder-decoder architecture that is trained to reconstruct images by disentangling the salient information of the image and the values of attributes directly in the latent space. As a result, after training, our model can generate different realistic versions of an input image by varying the attribute values. By using continuous attribute values, we can choose how much a specific attribute is perceivable in the generated image. This property could allow for applications where users can modify an image using sliding knobs, like faders on a mixing console, to change the facial expression of a portrait, or to update the color of some objects. Compared to the state-of-the-art which mostly relies on training adversarial networks in pixel space by altering attribute values at train time, our approach results in much simpler training schemes and nicely scales to multiple attributes. We present evidence that our model can significantly change the perceived value of the attributes while preserving the naturalness of images.

Recent work has explored deep architectures for learning acoustic features in an unsupervised or weakly-supervised way for phone recognition. Here we investigate the role of the input features, and in particular we test whether standard mel-scaled filterbanks could be replaced by inherently richer representations, such as derived from an analytic scattering spectrum. We use a Siamese network using lexical side information similar to a well-performing architecture used in the Zero Resource Speech Challenge (2015), and show a substantial improvement when the filterbanks are replaced by scattering features, even though these features yield similar performance when tested without training. This shows that unsupervised and weakly-supervised architectures can benefit from richer features than the traditional ones.