Speech Processing

In this study, we investigate the application of keyword spotting (KWS) in the domain of Hindi speech recognition, utilizing a dataset comprising 40,000 audio samples. With a sampling rate of 44 kHz and an average duration of 1.9 seconds per sample, we focus on developing an efficient on-device KWS system tailored for user-specific queries. Leveraging Convolutional Neural Networks (CNNs) for classification, we employ feature engineering techniques to convert raw audio recordings into Mel Frequency Cepstral Coefficients (MFCCs) as an input for our network. Our experiments encompass various CNN architectures, exploring their efficacy in identifying predefined keywords within the continuous speech stream. Our CNN-based approach achieves a commendable accuracy rate of 91.79% through rigorous evaluation, demonstrating promising performance while ensuring computational efficiency and user-specific customization in Hindi speech recognition.

Neural networks can be trained to learn task-relevant representations from data. Understanding how these networks make decisions falls within the Explainable AI (XAI) domain. This paper proposes to study an XAI topic: analysing, visualising and understanding the unknown organisation of network representations, particularly those a speaker recognition network learns from utterances, for recognising speaker identity. Past studies have employed algorithms (e.g. K-means) to analyse the different ways in which network representations can be naturally grouped into clusters, i.e. to analyse different flat clustering phenomena within the space defined by those representations. In contrast, this work applies two algorithms -- Single-Linkage Clustering (SLINK) and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) -- to analyse the different ways in which representations from the speaker recognition network can form clusters with hierarchical relationships, i.e., to analyse different hierarchical clustering phenomena within the representation space of the speaker recognition network. Furthermore, an algorithm called Hierarchical Cluster-Class Matching (HCCM) is designed to semantically interpret one of the above hierarchical clustering phenomena analysed using SLINK. Given the clusters representing this phenomenon, HCCM identifies which ones best match individual semantic classes related to gender and nationality (e.g.\ male, female, Ireland, UK) and and-logic conjunctions of these classes (e.g.\ female and Ireland). The Liebig score metric is also proposed within HCCM to quantify the matching quality of each cluster-class pair and diagnose the factor that limits each match.

In this work, we present Au-M-ol, a novel multimodal architecture that extends Large Language Models (LLMs) with audio processing. It is designed to improve performance on clinically relevant tasks such as Automatic Speech Recognition (ASR). Au-M-ol has three main components: (1) an audio encoder that extracts rich acoustic features from medical speech, (2) an adaptation layer that maps audio features into the LLM input space, and (3) a pretrained LLM that performs transcription and clinical language understanding. This design allows the model to interpret spoken medical content directly, improving both accuracy and robustness. In experiments, Au-M-ol reduces Word Error Rate (WER) by 56% compared to state-of-the-art baselines on medical transcription tasks. The model also performs well in challenging conditions, including noisy environments, domain-specific terminology, and speaker variability. These results suggest that Au-M-ol is a strong candidate for real-world clinical applications, where reliable and context-aware audio understanding is essential.

Human-imitated speech poses a greater challenge than AI-generated speech for both human listeners and automatic detection systems. Unlike AI-generated speech, which often contains artifacts, over-smoothed spectra, or robotic cues, imitated speech is produced naturally by humans, thereby preserving a higher degree of naturalness that makes imitation-based speech forgery significantly more challenging to detect using conventional acoustic or cepstral features. To overcome this challenge, this study proposes an auditory perception-based Spectro-Temporal Modulation (STM) representation framework for human-imitated speech detection. The STM representations are derived from two cochlear filterbank models: the Gammatone Filterbank (GTFB), which simulates frequency selectivity and can be regarded as a first approximation of cochlear filtering, and the Gammachirp Filterbank (GCFB), which further models both frequency selectivity and level-dependent asymmetry. These STM representations jointly capture temporal and spectral fluctuations in speech signals, corresponding to changes over time in the spectrogram and variations along the frequency axis related to human auditory perception. We also introduce a Segmental-STM representation to analyze short-term modulation patterns across overlapping time windows, enabling high-resolution modeling of temporal speech variations. Experimental results show that STM representations are effective for human-imitated speech detection, achieving accuracy levels close to those of human listeners. In addition, Segmental-STM representations are more effective, surpassing human perceptual performance. The findings demonstrate that perceptually inspired spectro-temporal modeling is promising for detecting imitation-based speech attacks and improving voice authentication robustness.

As part of the MediSpeech project, we aim to develop a system that transcribes and diarizes Dutch conversations between doctors and patients in real-time. In this research (in-progress) we explore ways of efficiently combining Whisper with speaker diarization (SD). After trying to prompt Whisper with text that contains speaker labels, we observed that it is able to insert labels into the transcription with promising accuracy. We continued this line of research by fine-tuning Whisper with speaker-labelled prompts to generate transcriptions in a format similar to that of Serialized Output Training (SOT). Fine-tuning Whisper yielded more consistent speaker IDs across the chunks of long-form audio and improved verbatim transcription. The study uncovered new challenges as Whisper's SD performance suffers because of mistakes that get propagated through prompts and inaccurate timestamps assigned to overlapping speech.

Mispronunciation Detection and Diagnosis (MDD) requires modeling fine-grained acoustic deviations. However, current ASR-derived MDD systems often face inherent limitations. In particular, CTC-based models favor sequence-level alignments that neglect transient mispronunciation cues, while explicit canonical priors bias predictions toward intended targets. To address these bottlenecks, we propose a prompt-free framework decoupling acoustic fidelity from canonical guidance. First, we introduce CROTTC, an acoustic model enforcing monotonic, frame-level alignment to accurately capture pronunciation deviations. Second, we implicitly inject mispronunciation information via the IF strategy under the knowledge transfer principle. Experiments show CROTTC-IF achieves a 71.77% F1-score on L2-ARCTIC and 71.70% F1-score on the Iqra'Eval2 leaderboard. With empirical analysis, we demonstrate that decoupling acoustics from explicit priors provides highly robust MDD.

Wav2vec 2.0 (W2V2) has shown strong performance in pathological speech analysis by effectively capturing the characteristics of atypical speech. Despite its success, it remains unclear which components of its learned representations are most informative for specific downstream tasks. In this study, we address this question by investigating the regression of dysarthric speech descriptors using annotations from the Speech Accessibility Project dataset. We focus on five descriptors, each addressing a different aspect of speech or voice production: intelligibility, imprecise consonants, inappropriate silences, harsh voice and monoloudness. Speech representations are derived from a W2V2-based feature extractor, and we systematically compare layer-wise and time-wise aggregation strategies using attentive statistics pooling. Our results show that intelligibility is best captured through layer-wise representations, whereas imprecise consonants, harsh voice and monoloudness benefit from time-wise modeling. For inappropriate silences, no clear advantage could be observed for either approach.

Atypical speech is receiving greater attention in speech technology research, but much of this work unfolds with limited interdisciplinary dialogue. For stuttered speech in particular, it is widely recognised that current speech recognition systems fall short in practice, and current evaluation methods and research priorities are not systematically grounded in end-user experiences and needs. In this work, we analyse these gaps through 1) a scoping review of papers that deal with stuttered speech and 2) a survey of 70 stakeholders, including adults who stutter and speech-language pathologists. By analysing these two perspectives, we propose a taxonomy of stuttered-speech research, identify where current research directions diverge from the needs articulated by stakeholders, and conclude by outlining concrete guidelines and directions towards addressing the real needs of the stuttering community.

Speaker verification is a task of confirming an individual's identity through the analysis of their voice. Whispered speech differs from phonated speech in acoustic characteristics, which degrades the performance of speaker verification systems in real-life scenarios, including avoiding fully phonated speech to protect privacy, disrupt others, or when the lack of full vocalization is dictated by a disease. In this paper we propose a model with a training recipe to obtain more robust representations against whispered speech hindrances. The proposed system employs an encoder--decoder structure built atop a fine-tuned speaker verification backbone, optimized jointly using cosine similarity--based classification and triplet loss. We gain relative improvement of 22.26% compared to the baseline (baseline 6.77% vs ours 5.27%) in normal vs whispered speech trials, achieving AUC of 98.16%. In tests comparing whispered to whispered, our model attains an EER of 1.88% with AUC equal to 99.73%, which represents a 15% relative enhancement over the prior leading ReDimNet-B2. We also offer a summary of the most popular and state-of-the-art speaker verification models in terms of their performance with whispered speech. Additionally, we evaluate how these models perform under noisy audios, obtaining that generally the same relative level of noise degrades the performance of speaker verification more significantly on whispered speech than on normal speech.

While generative models have set new benchmarks for Target Speaker Extraction (TSE), their inherent reliance on global context precludes deployment in real-time applications. Direct adaptation to streaming scenarios often leads to catastrophic inference performance degradation due to the severe mismatch between training and streaming inference. To bridge this gap, we present the first autoregressive (AR) models tailored for streaming TSE. Our approach introduces a Chunk-wise Interleaved Splicing Paradigm that ensures highly efficient and stable streaming inference. To ensure the coherence between the extracted speech segments, we design a historical context refinement mechanism that mitigates boundary discontinuities by leveraging historical information. Experiments on Libri2Mix show that while AR generative baseline exhibits performance degradation at low latencies, our approach maintains 100% stability and superior intelligibility. Furthermore, our streaming results are comparable to or even surpass offline baselines. Additionally, our model achieves a Real-Time-Factor (RTF) of 0.248 on consumer-level GPUs. This work provides empirical evidence that AR generative backbones are viable for latency-sensitive applications through the Chunk-wise Interleaved Splicing Paradigm.

Full-duplex speech interaction, as the most natural and intuitive mode of human communication, is driving artificial intelligence toward more human-like conversational systems. Traditional cascaded speech processing pipelines suffer from critical limitations, including accumulated latency, information loss, and error propagation across modules. To address these issues, recent efforts focus on the end-to-end audio large language models (LLMs) like GPT-4o, which primarily unify speech understanding and generation task. However, most of these models are inherently half-duplex, and rely on a suite of separate, task-specific front-end components, such as voice activity detection (VAD) and turn-taking detection (TD). In our development of speech assistant, we observed that optimizing the speech front-end is equally crucial as advancing the back-end unified model for achieving seamless, responsive interactions. To bridge this gap, we propose the first unified audio front-end LLM (UAF) tailored for full-duplex speech systems. Our model reformulates diverse audio front-end tasks into a single auto-regressive sequence prediction problem, including VAD, TD, speaker recognition (SR), automatic speech recognition (ASR) and question answer (QA). It takes streaming fixed-duration audio chunk (e.g., 600 ms) as input, leverages a reference audio prompt to anchor the target speaker at the beginning, and regressively generates discrete tokens encoding both semantic content and system-level state controls (e.g., interruption signals). Experiments demonstrate that our model achieves leading performance across multiple audio front-end tasks and significantly enhances response latency and interruption accuracy in real-world interaction scenarios.

Despite growing interest in Quranic data research, existing Quran datasets remain limited in both scale and diversity. To address this gap, we present Tadabur, a large-scale Quran audio dataset. Tadabur comprises more than 1400+ hours of recitation audio from over 600 distinct reciters, providing substantial variation in recitation styles, vocal characteristics, and recording conditions. This diversity makes Tadabur a comprehensive and representative resource for Quranic speech research and analysis. By significantly expanding both the total duration and variability of available Quran data, Tadabur aims to support future research and facilitate the development of standardized Quranic speech benchmarks.

We test whether Speech Articulatory Coding (SPARC) features can linearly predict surface electromyography (sEMG) envelopes across aloud, mimed, and subvocal speech in twenty-four subjects. Using elastic-net multivariate temporal response function (mTRF) with sentence-level cross-validation, SPARC yields higher prediction accuracy than phoneme one-hot representations on nearly all electrodes and in all speech modes. Aloud and mimed speech perform comparably, and subvocal speech remains above chance, indicating detectable articulatory activity. Variance partitioning shows a substantial unique contribution from SPARC and a minimal unique contribution from phoneme features. mTRF weight patterns reveal anatomically interpretable relationships between electrode sites and articulatory movements that remain consistent across modes. This study focuses on representation/encoding analysis (not end-to-end decoding) and supports SPARC as a robust and interpretable intermediate target for sEMG-based silent-speech modeling.

Speech encoder models are known to model members of some speaker groups (SGs) better than others. However, there has been little work in establishing why this occurs on a technological level. To our knowledge, we present the first layerwise fairness analysis of pretrained self-supervised speech encoder models (S3Ms), probing each embedding layer for speaker identification (SID) automatic speech recognition (ASR). We find S3Ms produce embeddings biased against certain SGs for both tasks, starting at the very first latent layers. Furthermore, we find opposite patterns of layerwise bias for SID vs ASR for all models in our study: SID bias is minimized in layers that minimize overall SID error; on the other hand, ASR bias is maximized in layers that minimize overall ASR error. The inverse bias/error relationship for ASR is unaffected when probing S3Ms that are finetuned for ASR, suggesting SG-level bias is established during pretraining and is difficult to remove.

Speech Emotion Recognition (SER) is the use of machines to detect the emotional state of humans based on the speech, which is gaining importance in natural human-computer interaction. Speech is a very valuable source of information, as emotions modify the patterns of speech; pitch, energy and even timing. Nonetheless, SER is not an easy task because speakers are not constant, and situations vary when recording and the sound similarity between specific feelings. In this work, the author introduces a speech emotion recognition system relying on the Mel-Frequency Cepstral Coefficient and Long Short-Term Memory (LSTM) neural network, as a feature extraction method. The Toronto Emotional Speech Set (TESS) speech signal was pre-processed, and transformed into MFCC features to understand the important aspects in terms of time. The resultant features were then introduced to LSTM model, which is able to learn long term features of sequential audio data. The trained model was measured over several emotion classes occurring in the dataset. As seen in the results of experiments, the proposed MFCC-LSTM approach succeeds in capturing the patterns of emotions in speech and provides highly realistic classifications in all the chosen emotion classifications. This study presents a speech emotion recognition system using Mel-Frequency Cepstral Coefficients (MFCCs) as features and a deep learning LSTM classifier. A Support Vector Machine (SVM) with an RBF kernel served as a classical baseline, achieving 98% accuracy, against which the proposed LSTM model, achieving 99% accuracy, was validated. Overall, it is possible to confirm that LSTM-based architectures can be used to address the task of speech emotion recognition. Actual applications of the proposed system may be virtual assistants and mental health surveillance.

In bandwidth-constrained communication such as satellite and underwater channels, speech must often be transmitted at ultra-low bitrates where intelligibility is the primary objective. At such extreme compression levels, codecs trained with acoustic reconstruction losses tend to allocate bits to perceptual detail, leading to substantial degradation in word error rate (WER). This paper proposes ClariCodec, a neural speech codec operating at 200 bit per second (bps) that reformulates quantisation as a stochastic policy, enabling reinforcement learning (RL)-based optimisation of intelligibility. Specifically, the encoder is fine-tuned using WER-driven rewards while the acoustic reconstruction pipeline remains frozen. Even without RL, ClariCodec achieves 3.68% WER on the LibriSpeech test-clean set at 200 bps, already competitive with codecs operating at higher bitrates. Further RL fine-tuning reduces WER to 3.20% on test-clean and 8.93% on test-other, corresponding to a 13% relative reduction while preserving perceptual quality.

In computational paralinguistics, detecting cognitive load and deception from speech signals is a heavily researched domain. Recent efforts have attempted to apply these acoustic frameworks to corporate earnings calls to predict catastrophic stock market volatility. In this study, we empirically investigate the limits of acoustic feature extraction (pitch, jitter, and hesitation) when applied to highly trained speakers in in-the-wild teleconference environments. Utilizing a two-stream late-fusion architecture, we contrast an acoustic-based stream with a baseline Natural Language Processing (NLP) stream. The isolated NLP model achieved a recall of 66.25% for tail-risk downside events. Surprisingly, integrating acoustic features via late fusion significantly degraded performance, reducing recall to 47.08%. We identify this degradation as Acoustic Camouflage, where media-trained vocal regulation introduces contradictory noise that disrupts multimodal meta-learners. We present these findings as a boundary condition for speech processing applications in high-stakes financial forecasting.

We introduce CONCORD, a privacy-aware asynchronous assistant-to-assistant (A2A) framework that leverages collaboration between proactive speech-based AI. As agents evolve from reactive to always-listening assistants, they face a core privacy risk (of capturing non-consenting speakers), which makes their social deployment a challenge. To overcome this, we implement CONCORD, which enforces owner-only speech capture via real-time speaker verification, producing a one-sided transcript that incurs missing context but preserves privacy. We demonstrate that CONCORD can safely recover necessary context through (1) spatio-temporal context resolution, (2) information gap detection, and (3) minimal A2A queries governed by a relationship-aware disclosure. Instead of hallucination-prone inferring, CONCORD treats context recovery as a negotiated safe exchange between assistants. Across a multi-domain dialogue dataset, CONCORD achieves 91.4% recall in gap detection, 96% relationship classification accuracy, and 97% true negative rate in privacy-sensitive disclosure decisions. By reframing always-listening AI as a coordination problem between privacy-preserving agents, CONCORD offers a practical path toward socially deployable proactive conversational agents.

Zero-shot voice conversion (VC) aims to convert a source utterance into the voice of an unseen target speaker while preserving its linguistic content. Although recent systems have improved conversion quality, building zero-shot VC systems for interactive scenarios remains challenging because high-fidelity speaker transfer and low-latency streaming inference are difficult to achieve simultaneously. In this work, we present X-VC, a zero-shot streaming VC system that performs one-step conversion in the latent space of a pretrained neural codec. X-VC uses a dual-conditioning acoustic converter that jointly models source codec latents and frame-level acoustic conditions derived from target reference speech, while injecting utterance-level target speaker information through adaptive normalization. To reduce the mismatch between training and inference, we train the model with generated paired data and a role-assignment strategy that combines standard, reconstruction, and reversed modes. For streaming inference, we further adopt a chunkwise inference scheme with overlap smoothing that is aligned with the segment-based training paradigm of the codec. Experiments on Seed-TTS-Eval show that X-VC achieves the best streaming WER in both English and Chinese, strong speaker similarity in same-language and cross-lingual settings, and substantially lower offline real-time factor than the compared baselines. These results suggest that codec-space one-step conversion is a practical approach for building high-quality low-latency zero-shot VC systems. Our audio samples, code and checkpoints are released at https://github.com/Jerrister/X-VC.

Natural language processing (NLP) and speech technologies have made significant progress in recent years; however, they remain largely focused on standardized language varieties. Dialects, despite their cultural significance and widespread use, are underrepresented in linguistic resources and computational models, resulting in performance disparities. To address this gap, we introduce Saar-Voice, a six-hour speech corpus for the Saarbrücken dialect of German. The dataset was created by first collecting text through digitized books and locally sourced materials. A subset of this text was recorded by nine speakers, and we conducted analyses on both the textual and speech components to assess the dataset's characteristics and quality. We discuss methodological challenges related to orthographic and speaker variation, and explore grapheme-to-phoneme (G2P) conversion. The resulting corpus provides aligned textual and audio representations. This serves as a foundation for future research on dialect-aware text-to-speech (TTS), particularly in low-resource scenarios, including zero-shot and few-shot model adaptation.

Voice imitation aims to transform source speech to match a reference speaker's timbre and speaking style while preserving linguistic content. A straightforward approach is to train on triplets of (source, reference, target), where source and target share the same content but target matches the reference's voice characteristics, yet such data is extremely scarce. Existing approaches either employ carefully designed disentanglement architectures to bypass this data scarcity or leverage external systems to synthesize pseudo-parallel training data. However, the former requires intricate model design, and the latter faces a quality ceiling when synthetic speech is used as training targets. To address these limitations, we propose MimicLM, which takes a novel approach by using synthetic speech as training sources while retaining real recordings as targets. This design enables the model to learn directly from real speech distributions, breaking the synthetic quality ceiling. Building on this data construction approach, we incorporate interleaved text-audio modeling to guide the generation of content-accurate speech and apply post-training with preference alignment to mitigate the inherent distributional mismatch when training on synthetic data. Experiments demonstrate that MimicLM achieves superior voice imitation quality with a simple yet effective architecture, significantly outperforming existing methods in naturalness while maintaining competitive similarity scores across speaker identity, accent, and emotion dimensions.

This survey provides a comprehensive catalog of publicly available text and speech resources for two West African languages: Hausa, an Afroasiatic language with approximately 80-100 million speakers, and Fongbe, a Niger-Congo language spoken by approximately 2 million people in Benin. These languages represent contrasting cases on the resource availability spectrum. We address the question: What is the current state of publicly available NLP resources for Hausa and Fongbe, and what gaps remain? Through systematic search of academic repositories, data platforms, and web sources, we catalog parallel corpora, monolingual text collections, speech datasets, pre-trained models, and evaluation benchmarks. For each resource, we document size, domain coverage, format, licensing, and accessibility. Our findings reveal that Hausa benefits from broader text resource diversity across news, encyclopedic, and educational domains. Fongbe, while having more limited text resources, has been the focus of recent academic speech data collection initiatives. Both languages are represented in Masakhane benchmarks for NER and POS tagging. We provide task-specific recommendations and identify priority gaps including domain-diverse Fongbe text and dedicated Hausa speech corpora.

Automatic Speech Recognition (ASR) is increasingly used in applications involving child speech, such as language learning and literacy acquisition. However, the effectiveness of such applications is limited by high ASR error rates. The negative effects can be mitigated by identifying in advance which ASR-outputs are reliable. This work aims to develop two novel approaches for selecting reliable ASR-output at the utterance level, one for selecting reliable read speech and one for dialogue speech material. Evaluations were done on an English and a Dutch dataset, each with a baseline and finetuned model. The results show that utterance-level selection methods for identifying reliably transcribed speech recordings have high precision for the best strategy (P > 97.4) for both read speech and dialogue material, for both languages. Using the current optimal strategy allows 21.0% to 55.9% of dialogue/read speech datasets to be automatically selected with low (UER of < 2.6) error rates.

Full-duplex dialogue audio, in which each speaker is recorded on a separate track, is an important resource for spoken dialogue research, but is difficult to collect at scale. Most in-the-wild two-speaker dialogue is available only as degraded monaural mixtures, making it unsuitable for systems requiring clean speaker-wise signals. We propose DialogueSidon, a model for joint restoration and separation of degraded monaural two-speaker dialogue audio. DialogueSidon combines a variational autoencoder (VAE) operates on the speech self-supervised learning (SSL) model feature, which compresses SSL model features into a compact latent space, with a diffusion-based latent predictor that recovers speaker-wise latent representations from the degraded mixture. Experiments on English, multilingual, and in-the-wild dialogue datasets show that DialogueSidon substantially improves intelligibility and separation quality over a baseline, while also achieving much faster inference.

Differentiable Digital Signal Processing (DDSP) pipelines for voice conversion rely on subtractive synthesis, where a periodic excitation signal is shaped by a learned spectral envelope to reconstruct the target voice. In DDSP-QbE, the excitation is generated via phase accumulation, producing a sawtooth-like waveform whose abrupt discontinuities introduce aliasing artefacts that manifest perceptually as buzziness and spectral distortion, particularly at higher fundamental frequencies. We propose two targeted improvements to the excitation stage of the DDSP-QbE subtractive synthesizer. First, we incorporate explicit voicing detection to gate the harmonic excitation, suppressing the periodic component in unvoiced regions and replacing it with filtered noise, thereby avoiding aliased harmonic content where it is most perceptually disruptive. Second, we apply Polynomial Band-Limited Step (PolyBLEP) correction to the phase-accumulated oscillator, substituting the hard waveform discontinuity at each phase wrap with a smooth polynomial residual that cancels alias-generating components without oversampling or spectral truncation. Together, these modifications yield a cleaner harmonic roll-off, reduced high-frequency artefacts, and improved perceptual naturalness, as measured by MOS. The proposed approach is lightweight, differentiable, and integrates seamlessly into the existing DDSP-QbE training pipeline with no additional learnable parameters.

We study device-addressed speech detection under pre-ASR edge deployment constraints, where systems must decide whether to forward audio before transcription under strict latency and compute limits. We show that, in multi-speaker environments with temporally ambiguous utterances, this task is more effectively modelled as a sequential routing problem over interaction history than as an utterance-local classification task. We formalize this as Sequential Device-Addressed Routing (SDAR) and present the Selective Attention System (SAS), an on-device implementation that instantiates this formulation. On a held-out 60-hour multi-speaker English test set, the primary audio-only configuration achieves F1=0.86 (precision=0.89, recall=0.83); with an optional camera, audio+video fusion raises F1 to 0.95 (precision=0.97, recall=0.93). Removing causal interaction history (Stage~3) reduced F1 from 0.95 to 0.57+/-0.03 in the audio+video configuration under our evaluation protocol. Among the tested components, this was the largest observed ablation effect, indicating that short-horizon interaction history carries substantial decision-relevant information in the evaluated setting. SAS runs fully on-device on ARM Cortex-A class hardware (<150 ms latency, <20 MB footprint). All results are from internal evaluation on a proprietary dataset evaluated primarily in English; a 5-hour evaluation subset may be shared for independent verification (Section 8.8).

Recent diffusion-based text-to-speech (TTS) models achieve high naturalness and expressiveness, yet often suffer from speaker drift, a subtle, gradual shift in perceived speaker identity within a single utterance. This underexplored phenomenon undermines the coherence of synthetic speech, especially in long-form or interactive settings. We introduce the first automatic framework for detecting speaker drift by formulating it as a binary classification task over utterance-level speaker consistency. Our method computes cosine similarity across overlapping segments of synthesized speech and prompts large language models (LLMs) with structured representations to assess drift. We provide theoretical guarantees for cosine-based drift detection and demonstrate that speaker embeddings exhibit meaningful geometric clustering on the unit sphere. To support evaluation, we construct a high-quality synthetic benchmark with human-validated speaker drift annotations. Experiments with multiple state-of-the-art LLMs confirm the viability of this embedding-to-reasoning pipeline. Our work establishes speaker drift as a standalone research problem and bridges geometric signal analysis with LLM-based perceptual reasoning in modern TTS.

Voice-based interfaces are widely used; however, achieving fair Wake-up Word detection across diverse speaker populations remains a critical challenge due to persistent demographic biases. This study evaluates the effectiveness of demographics-agnostic training techniques in mitigating performance disparities among speakers of varying sex, age, and accent. We utilize the OK Aura database for our experiments, employing a training methodology that excludes demographic labels, which are reserved for evaluation purposes. We explore (i) data augmentation techniques to enhance model generalization and (ii) knowledge distillation of pre-trained foundational speech models. The experimental results indicate that these demographics-agnostic training techniques markedly reduce demographic bias, leading to a more equitable performance profile across different speaker groups. Specifically, one of the evaluated techniques achieves a Predictive Disparity reduction of 39.94% for sex, 83.65% for age, and 40.48% for accent when compared to the baseline. This study highlights the effectiveness of label-agnostic methodologies in fostering fairness in Wake-up Word detection.

We introduce Full-Duplex-Bench-v3 (FDB-v3), a benchmark for evaluating spoken language models under naturalistic speech conditions and multi-step tool use. Unlike prior work, our dataset consists entirely of real human audio annotated for five disfluency categories, paired with scenarios requiring chained API calls across four task domains. We evaluate six model configurations -- GPT-Realtime, Gemini Live 2.5, Gemini Live 3.1, Grok, Ultravox v0.7, and a traditional Cascaded pipeline (Whisper$\rightarrow$GPT-4o$\rightarrow$TTS) -- across accuracy, latency, and turn-taking dimensions. GPT-Realtime leads on Pass@1 (0.600) and interruption avoidance (13.5%); Gemini Live 3.1 achieves the fastest latency (4.25s) but the lowest turn-take rate (78.0%); and the Cascaded baseline, despite a perfect turn-take rate, incurs the highest latency (10.12s). Across all systems, self-correction handling and multi-step reasoning under hard scenarios remain the most consistent failure modes.

Outbound AI calling systems must distinguish voicemail greetings from live human answers in real time to avoid wasted agent interactions and dropped calls. We present a lightweight approach that extracts 15 temporal features from the speech activity pattern of a pre-trained neural voice activity detector (VAD), then classifies with a shallow tree-based ensemble. Across two evaluation sets totaling 764 telephony recordings, the system achieves a combined 96.1% accuracy (734/764), with 99.3% (139/140) on an expert-labeled test set and 95.4% (595/624) on a held-out production set. In production validation over 77,000 calls, it maintained a 0.3% false positive rate and 1.3% false negative rate. End-to-end inference completes in 46 ms on a commodity dual-core CPU with no GPU, supporting 380+ concurrent WebSocket calls. In our search over 3,780 model, feature, and threshold combinations, feature importance was concentrated in three temporal variables. Adding transcription keywords or beep-based features did not improve the best real-time configuration and increased latency substantially. Our results suggest that temporal speech patterns are a strong signal for distinguishing voicemail greetings from live human answers.