We introduce LESS (Large Language Model Enhanced Semi-supervised Learning), a versatile framework that leverages Large Language Models (LLMs) to correct pseudo labels generated from in-the-wild data. Within the LESS framework, pseudo-labeled text from Automatic Speech Recognition (ASR) or Automatic Speech Translation (AST) of the unsupervised data is refined by an LLM, and augmented by a data filtering strategy to optimize LLM knowledge transfer efficiency. Experiments on both Mandarin ASR and Spanish-to-English AST tasks show that LESS achieves a notable absolute WER reduction of 3.77% on the Wenet Speech test set, as well as BLEU scores of 34.0 and 64.7 on Callhome and Fisher test sets respectively. These results validate the adaptability of LESS across different languages, tasks, and domains. Ablation studies conducted with various LLMs and prompt configurations provide novel insights into leveraging LLM-derived knowledge for speech processing applications.
We introduce a data-driven approach for enabling word-level timestamp prediction in the Canary model. Accurate timestamp information is crucial for a variety of downstream tasks such as speech content retrieval and timed subtitles. While traditional hybrid systems and end-to-end (E2E) models may employ external modules for timestamp prediction, our approach eliminates the need for separate alignment mechanisms. By leveraging the NeMo Forced Aligner (NFA) as a teacher model, we generate word-level timestamps and train the Canary model to predict timestamps directly. We introduce a new <|timestamp|> token, enabling the Canary model to predict start and end timestamps for each word. Our method demonstrates precision and recall rates between 80% and 90%, with timestamp prediction errors ranging from 20 to 120 ms across four languages, with minimal WER degradation. Additionally, we extend our system to automatic speech translation (AST) tasks, achieving timestamp prediction errors around 200 milliseconds.
Granite-speech LLMs are compact and efficient speech language models specifically designed for English ASR and automatic speech translation (AST). The models were trained by modality aligning the 2B and 8B parameter variants of granite-3.3-instruct to speech on publicly available open-source corpora containing audio inputs and text targets consisting of either human transcripts for ASR or automatically generated translations for AST. Comprehensive benchmarking shows that on English ASR, which was our primary focus, they outperform several competitors' models that were trained on orders of magnitude more proprietary data, and they keep pace on English-to-X AST for major European languages, Japanese, and Chinese. The speech-specific components are: a conformer acoustic encoder using block attention and self-conditioning trained with connectionist temporal classification, a windowed query-transformer speech modality adapter used to do temporal downsampling of the acoustic embeddings and map them to the LLM text embedding space, and LoRA adapters to further fine-tune the text LLM. Granite-speech-3.3 operates in two modes: in speech mode, it performs ASR and AST by activating the encoder, projector, and LoRA adapters; in text mode, it calls the underlying granite-3.3-instruct model directly (without LoRA), essentially preserving all the text LLM capabilities and safety. Both models are freely available on HuggingFace (https://huggingface.co/ibm-granite/granite-speech-3.3-2b and https://huggingface.co/ibm-granite/granite-speech-3.3-8b) and can be used for both research and commercial purposes under a permissive Apache 2.0 license.
Automatic Speech Translation (AST) datasets for Indian languages remain critically scarce, with public resources covering fewer than 10 of the 22 official languages. This scarcity has resulted in AST systems for Indian languages lagging far behind those available for high-resource languages like English. In this paper, we first evaluate the performance of widely-used AST systems on Indian languages, identifying notable performance gaps and challenges. Our findings show that while these systems perform adequately on read speech, they struggle significantly with spontaneous speech, including disfluencies like pauses and hesitations. Additionally, there is a striking absence of systems capable of accurately translating colloquial and informal language, a key aspect of everyday communication. To this end, we introduce BhasaAnuvaad, the largest publicly available dataset for AST involving 13 out of 22 scheduled Indian languages and English spanning over 44,400 hours and 17M text segments. BhasaAnuvaad contains data for English speech to Indic text, as well as Indic speech to English text. This dataset comprises three key categories: (1) Curated datasets from existing resources, (2) Large-scale web mining, and (3) Synthetic data generation. By offering this diverse and expansive dataset, we aim to bridge the resource gap and promote advancements in AST for Indian languages.
A rising interest in the modality extension of foundation language models warrants discussion on the most effective, and efficient, multimodal training approach. This work focuses on neural machine translation (NMT) and proposes a joint multimodal training regime of Speech-LLM to include automatic speech translation (AST). We investigate two different foundation model architectures, decoder-only GPT and encoder-decoder T5, extended with Canary-1B's speech encoder. To handle joint multimodal training, we propose a novel training framework called EMMeTT. EMMeTT improves training efficiency with the following: balanced sampling across languages, datasets, and modalities; efficient sequential data iteration; and a novel 2D bucketing scheme for multimodal data, complemented by a batch size optimizer (OOMptimizer). We show that a multimodal training consistently helps with both architectures. Moreover, SALM-T5 trained with EMMeTT retains the original NMT capability while outperforming AST baselines on four-language subsets of FLORES and FLEURS. The resultant Multimodal Translation Model produces strong text and speech translation results at the same time.
Large language models (LLMs) have demonstrated remarkable advancements in language understanding and generation. Building on the success of text-based LLMs, recent research has adapted these models to use speech embeddings for prompting, resulting in Speech-LLM models that exhibit strong performance in automatic speech recognition (ASR) and automatic speech translation (AST). In this work, we propose a novel approach to leverage ASR transcripts as prompts for AST in a Speech-LLM built on an encoder-decoder text LLM. The Speech-LLM model consists of a speech encoder and an encoder-decoder structure Megatron-T5. By first decoding speech to generate ASR transcripts and subsequently using these transcripts along with encoded speech for prompting, we guide the speech translation in a two-step process like chain-of-thought (CoT) prompting. Low-rank adaptation (LoRA) is used for the T5 LLM for model adaptation and shows superior performance to full model fine-tuning. Experimental results show that the proposed CoT prompting significantly improves AST performance, achieving an average increase of 2.4 BLEU points across 6 En->X or X->En AST tasks compared to speech prompting alone. Additionally, compared to a related CoT prediction method that predicts a concatenated sequence of ASR and AST transcripts, our method performs better by an average of 2 BLEU points.
Integrating audio encoders with LLMs through connectors has enabled these models to process and comprehend audio modalities, significantly enhancing speech-to-text tasks, including automatic speech recognition (ASR) and automatic speech translation (AST). However, these methods often overlook the critical aspect of language adaptation in multilingual settings, relying instead on multilingual data without adequately addressing language differences. To address this gap, we propose the Ideal-LLM model, which employs dual multilingual encoders to enrich language feature information and utilizes a language-adapted connector to target the adaptation of each language specifically. By leveraging the complementary strengths of Whisper and MMS encoders, our approach ensures richer multilingual representations. Additionally, the language-adapted connector enhances modal transformation via a language weight selector tailored for each language. Experimental results demonstrate that Ideal-LLM significantly improves ASR performance, achieving a 32.6% relative reduction in average word error rates compared to the standard speech encoder integrated with LLMs and yields an average BLEU score of 36.78 for AST task.
In this paper, we present MooER, a LLM-based large-scale automatic speech recognition (ASR) / automatic speech translation (AST) model of Moore Threads. A 5000h pseudo labeled dataset containing open source and self collected speech data is used for training. We achieve performance comparable to other open source models trained with up to hundreds of thousands of hours of labeled speech data. Meanwhile, experiments conducted on Covost2 Zh2en testset suggest that our model outperforms other open source Speech LLMs. A BLEU score of 25.2 can be obtained. The main contributions of this paper are summarized as follows. First, this paper presents a training strategy for encoders and LLMs on speech related tasks (including ASR and AST) using a small size of pseudo labeled data without any extra manual annotation and selection. Second, we release our ASR and AST models and plan to open-source our training code and strategy in the near future. Moreover, a model trained on 8wh scale training data is planned to be released later on.
Large-scale automatic speech translation systems today lack key features that help machine-mediated communication feel seamless when compared to human-to-human dialogue. In this work, we introduce a family of models that enable end-to-end expressive and multilingual translations in a streaming fashion. First, we contribute an improved version of the massively multilingual and multimodal SeamlessM4T model-SeamlessM4T v2. This newer model, incorporating an updated UnitY2 framework, was trained on more low-resource language data. SeamlessM4T v2 provides the foundation on which our next two models are initiated. SeamlessExpressive enables translation that preserves vocal styles and prosody. Compared to previous efforts in expressive speech research, our work addresses certain underexplored aspects of prosody, such as speech rate and pauses, while also preserving the style of one's voice. As for SeamlessStreaming, our model leverages the Efficient Monotonic Multihead Attention mechanism to generate low-latency target translations without waiting for complete source utterances. As the first of its kind, SeamlessStreaming enables simultaneous speech-to-speech/text translation for multiple source and target languages. To ensure that our models can be used safely and responsibly, we implemented the first known red-teaming effort for multimodal machine translation, a system for the detection and mitigation of added toxicity, a systematic evaluation of gender bias, and an inaudible localized watermarking mechanism designed to dampen the impact of deepfakes. Consequently, we bring major components from SeamlessExpressive and SeamlessStreaming together to form Seamless, the first publicly available system that unlocks expressive cross-lingual communication in real-time. The contributions to this work are publicly released and accessible at https://github.com/facebookresearch/seamless_communication
End-to-end automatic speech translation (AST) relies on data that combines audio inputs with text translation outputs. Previous work used existing large parallel corpora of transcriptions and translations in a knowledge distillation (KD) setup to distill a neural machine translation (NMT) into an AST student model. While KD allows using larger pretrained models, the reliance of previous KD approaches on manual audio transcripts in the data pipeline restricts the applicability of this framework to AST. We present an imitation learning approach where a teacher NMT system corrects the errors of an AST student without relying on manual transcripts. We show that the NMT teacher can recover from errors in automatic transcriptions and is able to correct erroneous translations of the AST student, leading to improvements of about 4 BLEU points over the standard AST end-to-end baseline on the English-German CoVoST-2 and MuST-C datasets, respectively. Code and data are publicly available.\footnote{\url{https://github.com/HubReb/imitkd_ast/releases/tag/v1.1}}
Research in multilingual speech-to-text translation is topical. Having a single model that supports multiple translation tasks is desirable. The goal of this work it to improve cross-lingual transfer learning in multilingual speech-to-text translation via semantic knowledge distillation. We show that by initializing the encoder of the encoder-decoder sequence-to-sequence translation model with SAMU-XLS-R, a multilingual speech transformer encoder trained using multi-modal (speech-text) semantic knowledge distillation, we achieve significantly better cross-lingual task knowledge transfer than the baseline XLS-R, a multilingual speech transformer encoder trained via self-supervised learning. We demonstrate the effectiveness of our approach on two popular datasets, namely, CoVoST-2 and Europarl. On the 21 translation tasks of the CoVoST-2 benchmark, we achieve an average improvement of 12.8 BLEU points over the baselines. In the zero-shot translation scenario, we achieve an average gain of 18.8 and 11.9 average BLEU points on unseen medium and low-resource languages. We make similar observations on Europarl speech translation benchmark.
Automatic speech translation is sensitive to speech recognition errors, but in a multilingual scenario, the same content may be available in various languages via simultaneous interpreting, dubbing or subtitling. In this paper, we hypothesize that leveraging multiple sources will improve translation quality if the sources complement one another in terms of correct information they contain. To this end, we first show that on a 10-hour ESIC corpus, the ASR errors in the original English speech and its simultaneous interpreting into German and Czech are mutually independent. We then use two sources, English and German, in a multi-source setting for translation into Czech to establish its robustness to ASR errors. Furthermore, we observe this robustness when translating both noisy sources together in a simultaneous translation setting. Our results show that multi-source neural machine translation has the potential to be useful in a real-time simultaneous translation setting, thereby motivating further investigation in this area.
We present Mu$^{2}$SLAM, a multilingual sequence-to-sequence model pre-trained jointly on unlabeled speech, unlabeled text and supervised data spanning Automatic Speech Recognition (ASR), Automatic Speech Translation (AST) and Machine Translation (MT), in over 100 languages. By leveraging a quantized representation of speech as a target, Mu$^{2}$SLAM trains the speech-text models with a sequence-to-sequence masked denoising objective similar to T5 on the decoder and a masked language modeling (MLM) objective on the encoder, for both unlabeled speech and text, while utilizing the supervised tasks to improve cross-lingual and cross-modal representation alignment within the model. On CoVoST AST, Mu$^{2}$SLAM establishes a new state-of-the-art for models trained on public datasets, improving on xx-en translation over the previous best by 1.9 BLEU points and on en-xx translation by 1.1 BLEU points. On Voxpopuli ASR, our model matches the performance of an mSLAM model fine-tuned with an RNN-T decoder, despite using a relatively weaker sequence-to-sequence architecture. On text understanding tasks, our model improves by more than 6\% over mSLAM on XNLI, getting closer to the performance of mT5 models of comparable capacity on XNLI and TydiQA, paving the way towards a single model for all speech and text understanding tasks.
Language barriers present a great challenge in our increasingly connected and global world. Especially within the medical domain, e.g. hospital or emergency room, communication difficulties and delays may lead to malpractice and non-optimal patient care. In the HYKIST project, we consider patient-physician communication, more specifically between a German-speaking physician and an Arabic- or Vietnamese-speaking patient. Currently, a doctor can call the Triaphon service to get assistance from an interpreter in order to help facilitate communication. The HYKIST goal is to support the usually non-professional bilingual interpreter with an automatic speech translation system to improve patient care and help overcome language barriers. In this work, we present our ASR system development efforts for this conversational telephone speech translation task in the medical domain for two languages pairs, data collection, various acoustic model architectures and dialect-induced difficulties.
In recent years, automatic speech-to-speech and speech-to-text translation has gained momentum thanks to advances in artificial intelligence, especially in the domains of speech recognition and machine translation. The quality of such applications is commonly tested with automatic metrics, such as BLEU, primarily with the goal of assessing improvements of releases or in the context of evaluation campaigns. However, little is known about how the output of such systems is perceived by end users or how they compare to human performances in similar communicative tasks. In this paper, we present the results of an experiment aimed at evaluating the quality of a real-time speech translation engine by comparing it to the performance of professional simultaneous interpreters. To do so, we adopt a framework developed for the assessment of human interpreters and use it to perform a manual evaluation on both human and machine performances. In our sample, we found better performance for the human interpreters in terms of intelligibility, while the machine performs slightly better in terms of informativeness. The limitations of the study and the possible enhancements of the chosen framework are discussed. Despite its intrinsic limitations, the use of this framework represents a first step towards a user-centric and communication-oriented methodology for evaluating real-time automatic speech translation.
In automatic speech translation (ST), traditional cascade approaches involving separate transcription and translation steps are giving ground to increasingly competitive and more robust direct solutions. In particular, by translating speech audio data without intermediate transcription, direct ST models are able to leverage and preserve essential information present in the input (e.g. speaker's vocal characteristics) that is otherwise lost in the cascade framework. Although such ability proved to be useful for gender translation, direct ST is nonetheless affected by gender bias just like its cascade counterpart, as well as machine translation and numerous other natural language processing applications. Moreover, direct ST systems that exclusively rely on vocal biometric features as a gender cue can be unsuitable and potentially harmful for certain users. Going beyond speech signals, in this paper we compare different approaches to inform direct ST models about the speaker's gender and test their ability to handle gender translation from English into Italian and French. To this aim, we manually annotated large datasets with speakers' gender information and used them for experiments reflecting different possible real-world scenarios. Our results show that gender-aware direct ST solutions can significantly outperform strong - but gender-unaware - direct ST models. In particular, the translation of gender-marked words can increase up to 30 points in accuracy while preserving overall translation quality.
We propose autoencoding speaker conversion for training data augmentation in automatic speech translation. This technique directly transforms an audio sequence, resulting in audio synthesized to resemble another speaker's voice. Our method compares favorably to SpecAugment on English$\to$French and English$\to$Romanian automatic speech translation (AST) tasks as well as on a low-resource English automatic speech recognition (ASR) task. Further, in ablations, we show the benefits of both quantity and diversity in augmented data. Finally, we show that we can combine our approach with augmentation by machine-translated transcripts to obtain a competitive end-to-end AST model that outperforms a very strong cascade model on an English$\to$French AST task. Our method is sufficiently general that it can be applied to other speech generation and analysis tasks.
For automatic speech translation (AST), end-to-end approaches are outperformed by cascaded models that transcribe with automatic speech recognition (ASR), then translate with machine translation (MT). A major cause of the performance gap is that, while existing AST corpora are small, massive datasets exist for both the ASR and MT subsystems. In this work, we evaluate several data augmentation and pretraining approaches for AST, by comparing all on the same datasets. Simple data augmentation by translating ASR transcripts proves most effective on the English--French augmented LibriSpeech dataset, closing the performance gap from 8.2 to 1.4 BLEU, compared to a very strong cascade that could directly utilize copious ASR and MT data. The same end-to-end approach plus fine-tuning closes the gap on the English--Romanian MuST-C dataset from 6.7 to 3.7 BLEU. In addition to these results, we present practical recommendations for augmentation and pretraining approaches. Finally, we decrease the performance gap to 0.01 BLEU using a Transformer-based architecture.
This paper investigates the influence of different acoustic features, audio-events based features and automatic speech translation based lexical features in complex emotion recognition such as curiosity. Pretrained networks, namely, AudioSet Net, VoxCeleb Net and Deep Speech Net trained extensively for different speech based applications are studied for this objective. Information from deep layers of these networks are considered as descriptors and encoded into feature vectors. Experimental results on the EmoReact dataset consisting of 8 complex emotions show the effectiveness, yielding highest F1 score of 0.85 as against the baseline of 0.69 in the literature.
We investigate end-to-end speech-to-text translation on a corpus of audiobooks specifically augmented for this task. Previous works investigated the extreme case where source language transcription is not available during learning nor decoding, but we also study a midway case where source language transcription is available at training time only. In this case, a single model is trained to decode source speech into target text in a single pass. Experimental results show that it is possible to train compact and efficient end-to-end speech translation models in this setup. We also distribute the corpus and hope that our speech translation baseline on this corpus will be challenged in the future.