Papers for

call center operators

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Contextual pictograms help fix speech transcription errors in French and Spanish

Contextual Pictograms for Repairing Multilingual ASR Errors

Abstract: Speech transcription errors can be difficult to recognize when competing word hypotheses are unfamiliar or visually similar. We describe an interface that combines alternative transcripts, replay and contextually checked pictograms for selected uncertain concepts. The system will initially support French and Spanish functional utterances. A counterbalanced study will assess whether pictorial feedback helps listeners recover the speaker's intended action beyond text alternatives alone. We will also quantify semantic mismatch and additional interaction time, since an inappropriate image can reinforce the wrong interpretation rather than repair it.

Tue 3 NovComputation and LanguageHuman-Computer Interaction
The gist
Sometimes, when computers transcribe speech, they make mistakes that are hard to spot because the wrong words look or sound alike. The authors created an interface that not only shows possible word alternatives but also uses pictures related to uncertain words to help users understand what was really said. They plan to test this system with French and Spanish phrases to see if images help people better guess the speaker’s intended meaning compared to text alone. They also want to measure whether confusing pictures slow down the process or make misunderstandings worse.
Open → 2611.05637v1

Interface Quilt creates patches to change app interfaces without code

Interface Quilt: Modifying Applications with Computer Use Agents and Generative UI

Abstract: Modifying an existing application is difficult when it does not provide programmatic access for that change. Recent advances in computer-use agents (CUAs) enable richer understanding and operation of existing interfaces, while generative UI (GenUI) makes it practical to generate new ones. We examine how these advances expand software modification through a design space of what can be changed and a framework for how modifications can be implemented. We then introduce Interface Quilt, a system that combines CUAs and GenUI to generate interface patches, which are new interfaces displayed over an existing application to modify its presentation, interaction, and capabilities. Through demonstrations across a variety of applications, we show that interface patches support a range of modifications without source code, APIs, or accessibility data.

Sun 1 NovHuman-Computer Interaction
The gist
Changing how an app looks or works is often hard because apps usually don’t let you directly change their code or settings. The authors show how computer-use agents—smart tools that understand app interfaces—and AI systems that can create new interfaces can be combined to patch apps from the outside. Their Interface Quilt system creates new interface layers on top of existing apps that can change their buttons, look, and features without needing the app’s original code or APIs. This approach works on many different apps and lets people modify software more easily.
Open → 2611.01891v1

Streaming system protects voice privacy and removes personal info fast

Toward Complete Speaker Anonymization: a Streaming Cascaded System for Voice Privacy and PII Removal

Abstract: We present a cascade streaming voice privacy system, consisting of ASR, NLP, and TTS modules. By leveraging recent improvements in all three aspects of speech and language processing, we show that such a system is capable of achieving a result comparable to state-of-the-art offline voice privacy systems, while providing controllable capabilities of content protection beyond voice. With an estimated utterance-onset end-to-end latency of 717 ms, the presented system preserves speech content with 11% WER on Librispeech, while detecting PII with a 69% SLU F1 score and achieving up to 47.66% EER against a semi-informed attacker.

Thu 29 OctSound
The gist
This paper shows a new system that can protect your voice identity when speaking, by removing personal information in real time. The system combines speech recognition, language understanding, and voice synthesis in a sequence to keep your spoken content clear while hiding who you are and sensitive details. The authors demonstrate it works almost as well as slower offline systems but with much less delay. They also provide ways to control how much content or personal info gets protected.
Open → 2610.39754v1

Voice anonymization keeps speech clear and identity safe

The Content Leash: Mitigating Degenerate Solutions in Invertible Speaker Anonymization

Abstract: Voice anonymization aims to protect a speaker's identity while preserving the linguistic content of speech, enabling voice data to be shared and processed with reduced privacy risk. This work focuses on reversible voice anonymization, where the original voice can be recovered by an authorized user rather than being permanently discarded. However, existing approaches can achieve apparent speaker anonymization by degrading the speech itself, while their effectiveness may also depend on the speaker-verification representation used during training. This creates a need for methods that protect speaker identity without sacrificing speech intelligibility and that remain effective against unseen speaker verifiers. We introduce content preserving training constraints to improve the privacy-utility trade-off by preventing anonymization from being achieved primarily through speech degradation. We further introduce a verifier-generalization strategy that incorporates a stronger speaker-verification representation during training, improving anonymization robustness beyond the verifier used in the original training setup. The resulting approach maintains intelligible speech, strengthens anonymization, and preserves reliable voice restoration, with evaluation on unseen speakers and held-out speaker verification models. On 38 unseen speakers, the approach raises the attacker's equal error rate from a measured floor of 2.8% to 26.3%, and to 27.9% against a held-out verifier, with 11.8% word error rate and 94.9% speaker similarity on key-based restoration. These improvements advance reversible voice anonymization toward practical privacy-preserving speech applications, where anonymized speech can be safely processed or shared while the original voice remains recoverable under authorized conditions.

Tue 27 OctSoundCryptography and Security
The gist
Protecting people’s voices when sharing speech data is important for privacy. This paper presents a method that hides who is speaking without making the speech hard to understand or ruining the original voice. The method also works well even against voice recognition tools it wasn’t specifically trained on. The authors show it can stop attackers from guessing the speaker while still letting authorized users restore the original voice clearly.
Open → 2610.36768v1

Speaker diarization improved for live and recorded audio

S2SND-v2: Unified Online and Offline Speaker Diarization with Speaker-Adapted w2v-BERT 2.0

Abstract: This paper presents S2SND-v2, a unified online and offline speaker diarization framework built on S2SND that improves acoustic feature extraction and the use of the resulting representations. For acoustic feature extraction, we introduce a speaker-adapted w2v-BERT 2.0 front end and compare full and pruned construction routes. To better use these features, overlap-aware activity-query masking improves speaker embedding extraction under overlap, while late-speaker training aligns unknown-speaker training with online inference conditions. Combining these improvements, the pruned system achieves competitive average diarization error rates of 14.52% online and 12.03% offline across eight test sets.

Mon 26 OctSound
The gist
Speaker diarization is the process of telling who spoke when in an audio recording or live conversation. This paper presents improvements to a system called S2SND-v2 that work both for online (live) and offline (recorded) scenarios. They enhance the system’s ability to understand speech features and handle overlapping voices more effectively. As a result, their method achieves competitive error rates on multiple test sets, meaning it can identify speakers more accurately in challenging conditions.
Open → 2610.34707v1

Speech enhancement improved by noisy target training with drifting loss

Noisy Target Training Meets Drifting for Weakly-Supervised Speech Enhancement

Abstract: In speech enhancement, standard supervised training requires pairs of noisy and clean speech. Such pairs are typically created by artificially degrading clean speech, leading to a domain mismatch between synthetic training data and real recordings. Noisy target training (NyTT) mitigates this limitation by training an enhancement model to predict a noisy recording from a noisier version generated by adding extra noise. However, since the training target is noisy, NyTT might learn to leave residual noise in the enhanced output. In this paper, we address this issue in NyTT by incorporating a Sinkhorn drifting loss that matches the distribution of the enhanced speech to that of unpaired clean speech. The Sinkhorn drifting loss attracts enhanced speech toward high-density regions of the clean-speech distribution while repelling it from the current enhanced-speech distribution in latent space. This loss uses only unpaired clean speech rather than the clean counterpart of each degraded recording, preserving the weakly-supervised nature of NyTT. Experimental results on VoiceBank-DEMAND and CHiME-3 real recordings demonstrate the efficacy of the proposed combination.

Fri 23 OctSound
The gist
Cleaning up noisy speech recordings usually requires matching noisy sounds with their clean versions, which can be hard to get from real life. The paper explains how the authors improve a method that uses only noisy sounds by teaching a computer to make noisy recordings less noisy using extra noise inputs. They add a special technique called a Sinkhorn drifting loss that helps the cleaned speech sound more like real clean speech without needing exact clean examples for training. Tests with real recordings show this approach works better at reducing noise.
Open → 2610.32870v1

Target speech extraction works better with music in the background

MuseFlow-TSE: Music-Aware Target Speech Extraction via Cross-Source Band-Splitting and Data-to-Data Flow Matching

Abstract: Target speech extraction (TSE) isolates a speaker's voice given an enrollment utterance, but existing work is validated almost exclusively without music. Discriminative methods are content-faithful yet masking-bounded; generative methods improve naturalness but may hallucinate, drift in identity, and require many steps. We present MuseFlow-TSE, a discriminative deletion, generative insertion framework for music-bearing scenes. Stage-1 is a music-aware cross-source band-split extractor that estimates target, interfering speakers, music, and noise under conservation and music-leakage constraints. Stage-2 is a data-to-data flow refiner initialized by the discriminative estimate, negatively conditioned on interference stems, and run for 1--4 difficulty-adaptive Euler steps with an end-to-end fidelity objective.

Wed 21 OctSound
The gist
Sometimes when people talk with music playing, it’s hard to pick out just one person's voice. This paper introduces a method that can separate a target speaker’s voice from other sounds including music, other people talking, and noise. The method works in two stages: first, it separates different sound sources including music, and then it improves the speech clarity using a refined process. The authors show their approach helps keep the speaker’s identity clear and makes the extraction more accurate even with music present.
Open → 2610.29119v1

Soloist separates voices from noisy and overlapping speech

Soloist: A Generative Model for Blind Voice Separation in the Wild

Abstract: Real-world voice separation must recover individual voices from overlapping speech, environmental noise, and music, yet isolated source recordings are rarely available for supervision. We introduce Soloist, a generative model for blind voice separation that autoregressively generates individual voice streams as audio-codec sequences conditioned on the mixture, without requiring a target-speaker query or a predefined voice count. We propose a two-stage training paradigm to learn from synthetic mixtures and adapt to real recordings. The first stage combines supervised separation on synthetic mixtures with isolated source targets and an auxiliary task that predicts speaker timelines from real recordings, learning who speaks when from natural acoustics. The second stage applies reinforcement learning to real recordings, with rewards supplied by a model judge that compares candidate separations based on voice purity and content preservation without isolated source targets. Across diverse scenarios, Soloist recovers clean speech from multi-speaker overlap and noisy recordings, and extracts singing vocals from music accompaniment while maintaining relatively high pitch accuracy and intelligibility. On real-world recordings with unknown voice counts, Soloist also separates voices well. Finally, we integrate Soloist into a full-duplex conversation processing pipeline that assembles separated outputs into continuous speaker tracks, demonstrating its practical utility in real conversations.

Wed 14 OctSound
The gist
Separating individual voices from recordings with many people talking over each other, background noise, and music is very hard, especially when you don’t know who or how many people are speaking. The authors introduce Soloist, a smart audio model that listens to mixed voices and recreates each speaker’s voice separately without needing prior information on the number of speakers. They teach Soloist in two phases using both made-up and real audio, improving its ability on real-world sounds. Soloist works well on conversations, noisy environments, and even music with singing, making it useful for processing real conversations accurately.
Open → 2610.19716v1

HINTT compares cascaded and unified speech models for multilingual diarization and transcription

HINTT Submission to the 2nd MLC-SLM Challenge: Comparing Cascaded and Unified Approaches to Diarization and ASR

Abstract: This paper presents the HINTT system submitted to the 2nd Challenge and Workshop on Multilingual Conversational Speech Language Model (MLC-SLM). We address multilingual speaker-attributed ASR, where systems must determine who spoke when and what was spoken. We investigate two modeling strategies for this problem: a cascaded pipeline that combines speaker diarization with speech-LLM-based ASR, and a unified speech LLM that directly generates speaker labels, timestamps, and transcriptions. Our final submission is based on the cascaded pipeline, consisting of a fine-tuned DiariZen diarization model, a fine-tuned Qwen3-ASR model, and LLM-based generative error correction. For comparison, we also fine-tune VibeVoice-ASR as a unified model using the same official training data. All task-specific fine-tuning and model selection are performed using only the official MLC-SLM data, without external data or pseudo-labels. Experimental results demonstrate that the cascaded system remains more reliable under the MLC-SLM Task 1 conditions, while unified speech LLMs offer a promising direction for future speaker-attributed ASR.

Tue 6 OctComputation and Language
The gist
Understanding who said what and when in a conversation across multiple languages is a tricky problem. The paper describes two ways to solve this: one that breaks the task into steps (finding speakers first, then transcribing), and another that tries to do everything at once with a single model. The authors find that, for this challenge, their step-by-step method works more reliably, but the all-in-one approach shows promise for the future.
Open → 2610.08063v1

Nonpreemptive scheduling learns job service speeds to cut queue length

Nonpreemptive Scheduling While Learning Context-Dependent Service Rates

Abstract: We study nonpreemptive contextual queueing bandits in a single-server system. Each job is represented by a $d$-dimensional context vector; in each round, a job may arrive with its context drawn from an unknown distribution $\mathcal{D}$, and its departure probability is determined by a logistic model of that context vector with an unknown parameter $θ^*$. The server learns from service outcomes while deciding which waiting job to serve and whether to idle, aiming to minimize queue-length regret, the gap between its expected terminal queue length and the minimum achievable by an admissible policy. Once selected, a job must be served until completion, and we refer to this as the nonpreemptive setting. A central challenge is that, even with full model knowledge, the optimal policy cannot in general be characterized by a simple myopic rule, since the optimal action can change with the remaining horizon at the same queue state. Nevertheless, when the model and horizon are known, the optimal action can be obtained through a finite-horizon Bellman recursion. Motivated by this, we propose Learn--Clear--Plan (LCP), which estimates the system and uses the resulting Bellman recursion to make horizon-dependent decisions. LCP achieves $\widetilde{O}(\sqrt{d/T})$ queue-length regret, while a lower-bound construction gives $Ω(\min\{1/\sqrt{d},\sqrt{d/T}\})$ regret for every learning policy on some instance, establishing optimality up to polylogarithmic factors when $T\ge d^2$. When the horizon is unknown, no horizon-independent policy achieves vanishing regret against the finite-horizon optimum. We therefore use SEPT, the policy that serves a waiting job with the highest probability of departure, as a fixed reference, and suggest an estimated-SEPT algorithm that achieves a tracking error of $\widetilde{O}(\sqrt{d/t})$ without knowing the model.

Tue 29 SeptMachine Learning
The gist
The paper looks at how a single-server system can schedule jobs that each come with information, or context, that affects how long they take to finish. The challenge is that once the server starts a job, it has to finish it before moving on, and the server doesn’t know in advance exactly how job features relate to finishing times. The authors propose a method that learns from past jobs and plans ahead to choose which job to serve next in order to keep queues short. They prove their method performs nearly as well as an optimal policy that already knows how jobs behave.
Open → 2609.37660v1

Speech deepfake detection improves using combined acoustic features

What Survives the Codec Shift: Pooled No-Vocals Residuals for Speech Deepfake Detection

Abstract: The transition from vocoder-based to neural-codec speech synthesis makes generalization more difficult for speech deepfake detectors, particularly those relying on speech-oriented representations. It remains unclear which acoustic representations retain discriminative information when the generation mechanism changes. We therefore compare 12 acoustic representations using a shared low-capacity linear classifier to identify effective evidence under codec shift. The analysis shows that hierarchical XLS-R leads on the pooled test set, while pooled no-vocals residual statistics perform best on the unseen-codec condition, revealing complementary behavior across generation conditions. Building on this finding, we propose MN-P, a dual-view detector that integrates an utterance-level pooled no-vocals representation with token-level XLS-R features through adaptive gating. The proposed MN-P reduces EER by 54.2% overall and by 60.9% on the codec-unseen condition relative to the best-performing retrained state-of-the-art system, with consistent gains across different detector backends. These results indicate that pooled no-vocals residual statistics provide effective complementary evidence for cross-generation speech deepfake detection.

Sun 27 SeptSoundArtificial IntelligenceMachine Learning
The gist
Detecting fake speech is getting harder because new ways to create it sound more realistic. The authors studied many sound features to find which ones best spot fakes across different speech generation methods. They found that a mix of two types of features, including one called no-vocals residuals, works especially well when the fake speech method is new or unseen. Combining these features into a single detector greatly reduced errors in spotting fake speech. This approach could help build stronger protections against audio deepfakes.
Open → 2609.33375v1

Speaker temporal memory improves multi speaker speech recognition accuracy

STAM-ASR: Speaker-Temporal Anchoring with Memory for Multi-Speaker ASR

Abstract: Natural conversations make both speech recognition and speaker attribution challenging for ASR, as speakers take turns, overlap, and reappear over time. We propose STAM-ASR, Speaker-Temporal Anchoring with Memory, a lightweight framework that extends an already pretrained AudioLLM for multi-speaker ASR. Without relying on an external diarization system, STAM-ASR learns speaker activity and speaker-aware representations directly from intermediate AudioLLM features. Hence providing explicit who and when cues to modulate the AudioLLM's semantic representation without explicit speech separation. STAM-ASR further maintains fixed-size speaker and conversational memories to carry complementary context across turns. We evaluate STAM-ASR on AMI, ICSI, LibriCSS, and NOTSOFAR-1 across close-talk, far-field, overlapping, and cross-domain conditions. Our reported results shows that speaker-temporal conditioning and memory provide complementary benefits, while the gap between reference and predicted speaker activity identifies robust speaker tracking as a key remaining challenge.

Thu 24 SeptSound
The gist
Understanding who is speaking and when they talk is hard in conversations where people speak over each other and come back later. The authors created a method called STAM-ASR that helps speech recognition systems figure out speaker identity and timing without needing separate steps to separate voices. It uses memory to remember who spoke earlier and builds on existing speech AI models. Tests on multiple real-world conversation types show that remembering speaker info and timing helps, but accurately tracking speakers over time remains tough.
Open → 2609.29805v1

Speech enhancement model improves transcription accuracy on real recordings

Transcript-Supervised Post-Training of Generative Speech Enhancement on Real Recordings via Reinforce Adjoint Matching

Abstract: We adapt Reinforce Adjoint Matching (RAM), a reward-based post-training method, to generative speech enhancement (SE). Starting from a pretrained SE model, RAM tilts the model's conditional distribution toward outputs with higher reward. During training, the current model generates enhanced speech on-policy, evaluates each generated endpoint with a potentially non-differentiable reward, and analytically re-noises the endpoint to construct inputs for a reward-guided regression objective. This enables post-training directly on real recordings using weak supervision, such as text transcripts, without requiring paired clean speech targets or reward gradients. We investigate word error rate (WER)-based post-training and whether recognition performance can be improved without compromising perceptual speech quality. Experiments on real CHiME-4 recordings reduce WER by 5.08 percentage points relative to pretrained FlowSE without reducing any of the reported non-intrusive speech quality metrics. A subjective listening test at the default reward scale finds no statistically significant preference between the post-trained and pretrained models.

Thu 24 SeptMachine Learning
The gist
Speech enhancement systems aim to improve the clarity of speech recorded in noisy environments. The authors show how to better train these systems using real noisy recordings and just their transcriptions, without needing clean reference audio. They adapt a technique called Reinforce Adjoint Matching to guide the model toward reducing speech recognition errors. Their improved model lowers the word error rate on real recordings without making the enhanced speech sound worse.
Open → 2609.29405v1

Multilingual challenge advances conversational speech recognition and understanding

The Second MLC-SLM Challenge: Multilingual Conversational Speech Diarization, Recognition, and Understanding

Abstract: This paper summarizes the Interspeech2026 second Multilingual Conversational Speech Language Model (MLC-SLM) Challenge, which aims to advance the development of effective multilingual conversational speech language models. We describe the two challenge tasks: multilingual conversational speech diarization and recognition, and multilingual conversational speech understanding, together with the released real-world conversational speech dataset, evaluation protocols, and baseline systems. The challenge attracted 91 teams worldwide, with 704 valid leaderboard results and 14 technical reports across the two tasks. Based on the participating systems, we summarize representative approaches and distill practical insights into multilingual conversational speech recognition and understanding to support future research in the community.

Wed 23 SeptSound
The gist
Figuring out who is speaking, what they say, and what it means is hard when people talk in multiple languages. This paper sums up a global challenge where teams worked on solving these problems using real conversations from everyday life. The challenge offered two main tasks: recognizing and separating speech by speakers in many languages, and understanding the meaning behind their speech. The authors provide details about the data, rules, and starting systems, and share what they learned from the best teams to help future work.
Open → 2609.27514v1

Small language model quickly detects scam calls with calibrated confidence

Open-Jev Judgments on CallScreenBench: Calibrated One-Pass Scam Screening with a Small Language Model

Abstract: Screening a phone call for fraud needs a trustworthy probability after every caller turn, in milliseconds. Jev-style typed decisions promise exactly that: declared options go in, one calibrated probability per option comes out of a single forward pass, with no generated text. We test an open implementation of this readout, JevLite, on scam-call screening: Qwen3-4B is LoRA-tuned so that the temperature-scaled softmax over two answer-label logits is P(scam). On 41 held-out CallScreenBench scenarios (577 per-turn decisions) a three-seed ensemble reaches AUROC .974 with calibration error .052, non-inferior to an LLM judge (MiniMax-M3) at a pre-registered .02 margin, with no false alarms on legitimate calls, decisions 1.14 turns earlier under the same hang-up rule, and 64.5 ms per decision on one consumer GPU, 4.9x lower than the same backbone fine-tuned to generate its answer. The gain is in the readout and calibration, not accuracy: a fine-tuned ModernBERT encoder is not significantly worse, the recipe was selected with test-set exposure, and all callers are synthetic. We claim no architectural novelty; the contribution is the application and an evaluation reporting calibration, false alarms and decision timing alongside AUROC.

Mon 21 SeptComputation and Language
The gist
Phone calls can sometimes be scams, and it's important to know quickly if a call is risky. The authors tested a way to use a small language model to decide if a call is a scam right after each part of the conversation. This method gives a trustworthy probability instead of a generated text answer, which makes decisions faster. They found it works about as well as bigger models but much faster and with fewer false alarms on safe calls.
Open → 2609.23959v1

Style aware paraphrasing reduces text authorship identification risks

I Am No One: Style-Aware Paraphrasing for Text Anonymization

Abstract: Authorship attribution models can re-identify users from seemingly anonymized text by exploiting stable stylistic fingerprints, even after explicit identifiers are removed, posing a growing privacy risk for text publishing and analytics. This risk extends to speech-derived text such as ASR transcripts of meetings and call-center conversations, where stylometric leakage can persist even after acoustic anonymization. Differential privacy-based anonymization often severely degrades text quality and utility. We propose a style-aware, prompt-driven anonymization approach that uses pretrained large language models to construct compact stylistic profiles from minimal samples and rewrite text to suppress identifiable style markers while preserving meaning. Across blog and review datasets, our approach reduces authorship attribution F1 by 60-70% while maintaining content quality and readability, substantially outperforming DP-based and non-DP baselines.

Fri 11 SeptComputation and LanguageCryptography and Security
The gist
Text can reveal who wrote it based on writing style even if names are removed. The authors found a way to rewrite text using AI that changes the style to hide who wrote it, while keeping the meaning clear. This method makes it much harder to guess the author and keeps the text easy to read. It works better than past privacy methods that often made texts hard to understand.
Open → 2609.12341v1

Qwen audio 3.0 ASR system recognizes diverse languages and dialects

Qwen-Audio-3.0-ASR Technical Report

Abstract: In recent years, automatic speech recognition (ASR) has witnessed transformative advancements driven by three complementary paradigms: data scaling, model scaling, and deep integration with large language models (LLMs). However, bridging the gap between academic benchmark performance and real-world production utility remains a persistent challenge, particularly in handling diverse regional dialects, dynamic entities and hotwords, long-range contextual information, and disfluent spontaneous speech. In this report, we present Qwen-Audio-3.0-ASR, a Mixture-of-Experts (MoE) LLM-based ASR system designed to address these production demands through a unified, instruction-following framework. The model is built upon the Qwen backbone, and is trained on tens of millions of hours of large-scale speech data. Qwen-Audio-3.0-ASR supports transcription across 30 languages and 16 Chinese dialectal varieties spanning eight major dialect regions. Beyond multilingual and dialectal recognition, the model provides production-oriented capabilities including industry-domain entity recognition, hierarchical hotword customization, native single-pass transcription polishing, and long-audio contextual modeling. We further develop a dedicated streaming variant, Qwen-Audio-3.0-ASR-Streaming, for latency-sensitive applications. Extensive evaluations on Chinese, English, multilingual, and real-world industrial test sets demonstrate state-of-the-art or highly competitive recognition performance across a broad range of evaluation conditions, with strong performance relative to leading commercial and proprietary systems including GPT-4o Transcribe and Gemini 3.1 Pro.

Mon 7 SeptComputation and Language
The gist
Speech recognition technology can struggle with different accents, long conversations, and unusual words. The authors created Qwen-Audio-3.0-ASR, a speech recognition system that understands 30 languages and 16 Chinese dialects. It handles long audio, special words, and spontaneous speech better by combining a large language model with expert modules. This system also includes a fast streaming version for real-time transcription needs. Tests show it works as well or better than top commercial products in many real-world situations.
Open → 2609.07549v1