Similar to text-based Large Language Models (LLMs), Speech-LLMs exhibit emergent abilities and context awareness. However, whether these similarities extend to gender bias remains an open question. This study proposes a methodology leveraging speaker assignment as an analytic tool for bias investigation. Unlike text-based models, which encode gendered associations implicitly, Speech-LLMs must produce a gendered voice, making speaker selection an explicit bias cue. We evaluate Bark, a Text-to-Speech (TTS) model, analyzing its default speaker assignments for textual prompts. If Bark's speaker selection systematically aligns with gendered associations, it may reveal patterns in its training data or model design. To test this, we construct two datasets: (i) Professions, containing gender-stereotyped occupations, and (ii) Gender-Colored Words, featuring gendered connotations. While Bark does not exhibit systematic bias, it demonstrates gender awareness and has some gender inclinations.

Recent advancements in spontaneous text-to-speech (TTS) have enabled the realistic synthesis of creaky voice, a voice quality known for its diverse pragmatic and paralinguistic functions. In this study, we used synthesized creaky voice in perceptual tests, to explore how listeners without formal training perceive two distinct types of creaky voice. We annotated a spontaneous speech corpus using creaky voice detection tools and modified a neural TTS engine with a creaky phonation embedding to control the presence of creaky phonation in the synthesized speech. We performed an objective analysis using a creak detection tool which revealed significant differences in creaky phonation levels between the two creaky voice types and modal voice. Two subjective listening experiments were performed to investigate the effect of creaky voice on perceived certainty, valence, sarcasm, and turn finality. Participants rated non-positional creak as less certain, less positive, and more indicative of turn finality, while positional creak was rated significantly more turn final compared to modal phonation.

Recent advances in generative language modeling applied to discrete speech tokens presented a new avenue for text-to-speech (TTS) synthesis. These speech language models (SLMs), similarly to their textual counterparts, are scalable, probabilistic, and context-aware. While they can produce diverse and natural outputs, they sometimes face issues such as unintelligibility and the inclusion of non-speech noises or hallucination. As the adoption of this innovative paradigm in speech synthesis increases, there is a clear need for an in-depth evaluation of its capabilities and limitations. In this paper, we evaluate TTS from a discrete token-based SLM, through both automatic metrics and listening tests. We examine five key dimensions: speaking style, intelligibility, speaker consistency, prosodic variation, spontaneous behaviour. Our results highlight the model’s strength in generating varied prosody and spontaneous outputs. It is also rated higher in naturalness and context appropriateness in listening tests compared to a conventional TTS. However, the model’s performance in intelligibility and speaker consistency lags behind traditional TTS. Additionally, we show that increasing the scale of SLMs offers a modest boost in robustness. Our findings aim to serve as a benchmark for future advancements in generative SLMs for speech synthesis.

As text-to-speech technologies achieve remarkable naturalness in read-aloud tasks, there is growing interest in multimodal synthesis of verbal and non-verbal communicative behaviour, such as spontaneous speech and associated body gestures. This paper presents a novel, unified architecture for jointly synthesising speech acoustics and skeleton-based 3D gesture motion from text, trained using optimal-transport conditional flow matching (OT-CFM). The proposed architecture is simpler than the previous state of the art, has a smaller memory footprint, and can capture the joint distribution of speech and gestures, generating both modalities together in one single process. The new training regime, meanwhile, enables better synthesis quality in much fewer steps (network evaluations) than before. Uni- and multimodal subjective tests demonstrate improved speech naturalness, gesture human-likeness, and cross-modal appropriateness compared to existing benchmarks.

We introduce Matcha-TTS, a new encoder-decoder architecture for speedy TTS acoustic modelling, trained using optimal-transport conditional flow matching (OT-CFM). This yields an ODE-based decoder capable of high output quality in fewer synthesis steps than models trained using score matching. Careful design choices additionally ensure each synthesis step is fast to run. The method is probabilistic, non-autoregressive, and learns to speak from scratch without external alignments. Compared to strong pre-trained baseline models, the Matcha-TTS system has the smallest memory footprint, rivals the speed of the fastest model on long utterances, and attains the highest mean opinion score in a listening test.

Mainstream text-to-speech (TTS) technologies predominantly rely on binary, cisgender speech, failing to adequately represent the diversity of gender expansive (e.g., transgender and/or nonbinary) people. This poses challenges, particularly for users of Speech Generating Devices (SGDs) seeking TTS voices that authentically reflect their identity and desired expressive nuances. This paper introduces a novel approach for constructing a palette of controllable gender-expansive TTS voices using recordings from 14 gender-expansive speakers. We employ Constrained PCA to extract gender-independent speaker identity vectors from x-vectors, using acoustic Vocal Tract Length (aVTL) as a known component.The result is applied as a speaker embedding in neural TTS, allowing control over the aVTL and several emergent properties captured as a representation of the vocal space across speakers. In addition to quantitative metrics, we present a community evaluation conducted by nonbinary SGD users.

As conversational Text-to-Speech (TTS) technologies become increasingly realistic and expressive, understanding the impact of prosodic variation on speech perception and social dynamics is crucial for enhancing conversational systems.This study explores the influence of prosodic features on listener responses to indirect requests using a specifically designed conversational TTS engine capable of controlling prosody, and generating speech across three different speaker profiles: female, male, and gender-ambiguous.We conducted two experiments to analyse how naturalistic variations in speech rate and vocal effort impact the likelihood of request compliance and perceived politeness.In the first experiment, we examined how prosodic modifications affect the perception of politeness in permission-and action requests.In the second experiment participants compared pairs of spoken requests, each rendered with different prosodic features, and chose which they were more likely to grant.Results indicate that both faster speech rate and higher vocal effort increased the willingness to comply, though the extent of this influence varied by speaker gender.Higher vocal effort in action requests increases the chance of being granted more than in permission requests.Politeness has a demonstrated positive impact on the likelihood of requests being granted, this effect is stronger for the male voice compared to female and gender-ambiguous voices.

Embodied conversational agents and social robots need to be able to generate spontaneous behavior in order to be believable in social interactions. We present a system that can generate spontaneous speech with supporting lip movements. The conversational TTS voice is trained on a podcast corpus that has been prosodically tagged (f0, speaking rate and energy) and transcribed (including tokens for breathing, fillers and laughter). We introduce a speech animation algorithm where articulatory effort can be adjusted. The speech animation is driven by time-stamped phonemes obtained from the internal alignment attention map of the TTS system, and we use prominence estimates from the synthesised speech waveform to modulate the lip- and jaw movements accordingly.

The emphasis on one-to-one speak/wait spoken conversational interaction with intelligent agents leads to long pauses between conversational turns, undermines the flow and naturalness of the interaction, and undermines the user experience. Despite ground breaking advances in the area of generating and understanding natural language with techniques such as LLMs, conversational interaction has remained relatively overlooked. In this workshop we will discuss and review the challenges, recent work and potential impact of improving conversational interaction with artificial systems. We hope to share experiences of poor human/system interaction, best practices with third party tools, and generate design guidance for the community.

Spontaneous speech has many affective and pragmatic functions that are interesting and challenging to model in TTS. However, the presence of reduced articulation, fillers, repetitions, and other disfluencies in spontaneous speech make the text and acoustics less aligned than in read speech, which is problematic for attention-based TTS. We propose a TTS architecture that can rapidly learn to speak from small and irregular datasets, while also reproducing the diversity of expressive phenomena present in spontaneous speech. Specifically, we add utterance-level prosody control to an existing neural HMM-based TTS system which is capable of stable, monotonic alignments for spontaneous speech. We objectively evaluate control accuracy and perform perceptual tests that demonstrate that prosody control does not degrade synthesis quality. To exemplify the power of combining prosody control and ecologically valid data for reproducing intricate spontaneous speech phenomena, we evaluate the system’s capability of synthesizing two types of creaky voice.

Data and code for "Synthesis after a couple PINTs: Investigating the role of pause-internal phonetic particles in speech synthesis and perception" at Interspeech 2023, Dublin, Aug 20-24.

Neural HMMs are a type of neural transducer recently proposed for sequence-to-sequence modelling in text-to-speech.They combine the best features of classic statistical speech synthesis and modern neural TTS, requiring less data and fewer training updates, and are less prone to gibberish output caused by neural attention failures.In this paper, we combine neural HMM TTS with normalising flows for describing the highly non-Gaussian distribution of speech acoustics.The result is a powerful, fully probabilistic model of durations and acoustics that can be trained using exact maximum likelihood.Experiments show that a system based on our proposal needs fewer updates than comparable methods to produce accurate pronunciations and a subjective speech quality close to natural speech.

Engaging embodied conversational agents need to generate expressive behavior in order to be believable in socializing interactions. We present a system that can generate spontaneous speech with supporting lip movements. The neural conversational TTS voice is trained on a multi-style speech corpus that has been prosodically tagged (pitch and speaking rate) and transcribed (including tokens for breathing, fillers and laughter). We introduce a speech animation algorithm where articulatory effort can be adjusted. The facial animation is driven by time-stamped phonemes and prominence estimates from the synthesised speech waveform to modulate the lip-and jaw movements accordingly. In objective evaluations we show that the system is able to generate speech and facial animation that vary in articulation effort. In subjective evaluations we compare our conversational TTS system's capability to deliver jokes with a commercial TTS. Both system succeeded equally good.

Many robots use their voice to communicate with people in spoken language but the voices commonly used for robots are often optimized for transactional interactions, rather than social ones. This can limit their ability to create engaging and natural interactions. To address this issue, we designed a spontaneous text-to-speech tool and used it to author natural and spontaneous robot speech. A crowdsourcing evaluation methodology is proposed to compare this type of speech to natural speech and state-of-the-art text-to-speech technology, both in disembodied and embodied form. We created speech samples in a naturalistic setting of people playing tabletop games and conducted a user study evaluating Naturalness, Intelligibility, Social Impression, Prosody, and Perceived Intelligence. The speech samples were chosen to represent three contexts that are common in tabletop games and the contexts were introduced to the participants that evaluated the speech samples. The study results show that the proposed evaluation methodology allowed for a robust analysis that successfully compared the different conditions. Moreover, the spontaneous voice met our target design goal of being perceived as more natural than a leading commercial text-to-speech.

When deploying robots, its physical characteristics, role, and tasks are often fixed. Such factors can also be associated with gender stereotypes among humans, which then transfer to the robots. One factor that can induce gendering but is comparatively easy to change is the robot’s voice. Designing voice in a way that interferes with fixed factors might therefore be a way to reduce gender stereotypes in human-robot interaction contexts. To this end, we have conducted a video-based online study to investigate how factors that might inspire gendering of a robot interact. In particular, we investigated how giving the robot a gender-ambiguous voice can affect perception of the robot. We compared assessments (n=111) of videos in which a robot’s body presentation and occupation mis/matched with human gender stereotypes. We found evidence that a gender-ambiguous voice can reduce gendering of a robot endowed with stereotypically feminine or masculine attributes. The results can inform more just robot design while opening new questions regarding the phenomenon of robot gendering.

Neural sequence-to-sequence TTS has achieved significantly better output quality than statistical speech synthesis using HMMs. However, neural TTS is generally not probabilistic and uses non-monotonic attention. Attention failures increase training time and can make synthesis babble incoherently. This paper describes how the old and new paradigms can be combined to obtain the advantages of both worlds, by replacing attention in neural TTS with an autoregressive left-right no-skip hidden Markov model defined by a neural network. Based on this proposal, we modify Tacotron 2 to obtain an HMM-based neural TTS model with monotonic alignment, trained to maximise the full sequence likelihood without approximation. We also describe how to combine ideas from classical and contemporary TTS for best results. The resulting example system is smaller and simpler than Tacotron 2, and learns to speak with fewer iterations and less data, whilst achieving comparable naturalness prior to the post-net. Our approach also allows easy control over speaking rate.

Inserting fillers (such as “um”, “like”) to clean speech text has a rich history of study. One major application is to make dialogue systems sound more spontaneous. The ambiguity of filler occurrence and inter-speaker difference make both modeling and evaluation difficult. In this paper, we study sampling-based filler insertion, a simple yet unexplored approach to inserting fillers. We propose an objective score called Filler Perplexity (FPP). We build three models trained on two single-speaker spontaneous corpora, and evaluate them with FPP and perceptual tests. We implement two innovations in perceptual tests, (1) evaluating filler insertion on dialogue systems output, (2) synthesizing speech with neural spontaneous TTS engines. FPP proves to be useful in analysis but does not correlate well with perceptual MOS. Perceptual results show little difference between compared filler insertion models including with ground-truth, which may be due to the ambiguity of what is good filler insertion and a strong neural spontaneous TTS that produces natural speech irrespective of input. Results also show preference for filler-inserted speech synthesized with spontaneous TTS. The same test using TTS based on read speech obtains the opposite results, which shows the importance of using spontaneous TTS in evaluating filler insertions. Audio samples: www.speech.kth.se/tts-demos/LREC22

Text-to-speech and co-speech gesture synthesis have until now been treated as separate areas by two different research communities, and applications merely stack the two technologies using a simple system-level pipeline. This can lead to modeling inefficiencies and may introduce inconsistencies that limit the achievable naturalness. We propose to instead synthesize the two modalities in a single model, a new problem we call integrated speech and gesture synthesis (ISG). We also propose a set of models modified from state-of-the-art neural speech-synthesis engines to achieve this goal. We evaluate the models in three carefully-designed user studies, two of which evaluate the synthesized speech and gesture in isolation, plus a combined study that evaluates the models like they will be used in real-world applications – speech and gesture presented together. The results show that participants rate one of the proposed integrated synthesis models as being as good as the state-of-the-art pipeline system we compare against, in all three tests. The model is able to achieve this with faster synthesis time and greatly reduced parameter count compared to the pipeline system, illustrating some of the potential benefits of treating speech and gesture synthesis together as a single, unified problem.

Breathing and speech planning in spontaneous speech are coordinated processes, often exhibiting disfluent patterns. While synthetic speech is not subject to respiratory needs, integrating breath into synthesis has advantages for naturalness and recall. At the same time, a synthetic voice reproducing disfluent breathing patterns learned from the data can be problematic. To address this, we first propose training stochastic TTS on a corpus of overlapping breath-group bigrams, to take context into account. Next, we introduce an unsupervised automatic annotation of likely-disfluent breath events, through a product-of-experts model that combines the output of two breath- event predictors, each using complementary information and operating in opposite directions. This annotation enables creating an automatically-breathing spontaneous speech synthesiser with a more fluent breathing style. A subjective evaluation on two spoken genres (impromptu and rehearsed) found the proposed system to be preferred over the baseline approach treating all breath events the same.

Embodied human communication encompasses both verbal (speech) and non-verbal information (e.g., gesture and head movements). Recent advances in machine learning have substantially improved the technologies for generating synthetic versions of both of these types of data: On the speech side, text-to-speech systems are now able to generate highly convincing, spontaneous-sounding speech using unscripted speech audio as the source material. On the motion side, probabilistic motion-generation methods can now synthesise vivid and lifelike speech-driven 3D gesticulation. In this paper, we put these two state-of-the-art technologies together in a coherent fashion for the first time. Concretely, we demonstrate a proof-of-concept system trained on a single-speaker audio and motion-capture dataset, that is able to generate both speech and full-body gestures together from text input. In contrast to previous approaches for joint speech-and-gesture generation, we generate full-body gestures from speech synthesis trained on recordings of spontaneous speech from the same person as the motion-capture data. We illustrate our results by visualising gesture spaces and textspeech-gesture alignments, and through a demonstration video.

By definition, spontaneous speech is unscripted and created on the fly by the speaker. It is dramatically different from read speech, where the words are authored as text before they are spoken. Spontaneous speech is emergent and transient, whereas text read out loud is pre-planned. For this reason, it is unsuitable to evaluate the usability and appropriateness of spontaneous speech synthesis by having it read out written texts sampled from for example newspapers or books. Instead, we need to use transcriptions of speech as the target - something that is much less readily available. In this paper, we introduce Starmap, a tool allowing developers to select a varied, representative set of utterances from a spoken genre, to be used for evaluation of TTS for a given domain. The selection can be done from any speech recording, without the need for transcription. The tool uses interactive visualisation of prosodic features with t-SNE, along with a tree-based algorithm to guide the user through thousands of utterances and ensure coverage of a variety of prompts. A listening test has shown that with a selection of genre-specific utterances, it is possible to show significant differences across genres between two synthetic voices built from spontaneous speech.

This paper considers utilising breaths to create improved spontaneous-speech corpora for conversational text-to-speech from found audio recordings such as dialogue podcasts. Breaths are of interest since they relate to prosody and speech planning and are independent of language and transcription. Specifically, we propose a semi-supervised approach where a fraction of coarsely annotated data is used to train a convolutional and recurrent speaker-specific breath detector operating on spectrograms and zero-crossing rate. The classifier output is used to find target-speaker breath groups (audio segments delineated by breaths) and subsequently select those that constitute clean utterances appropriate for a synthesis corpus. An application to 11 hours of raw podcast audio extracts 1969 utterances (106 minutes), 87% of which are clean and correctly segmented. This outperforms a baseline that performs integrated VAD and speaker attribution without accounting for breaths.

Betz S, Zarrieß S, Székely É, Wagner P. The greennn tree - lengthening position influences uncertainty perception. In: Proceedings of Interspeech. 2019: 3990-3994.

Synthesising spontaneous speech is a difficult task due to disfluencies, high variability and syntactic conventions different from those of written language. Using found data, as opposed to lab-rec ...

As synthetic voices become more flexible, and conversational systems gain more potential to adapt to the environmental and social situation, the question needs to be examined, how different modific ...

Conversation partners have a tendency to adapt their vocal in- tensity to each other and to other social and environmental fac- tors. A socially adequate vocal intensity level by a speech syn- thesiser that goes beyond mere volume adjustment is highly de- sirable for a rewarding and successful human-machine or ma- chine mediated human-human interaction. This paper examines the interaction of the Lombard effect and speaker entrainment in a controlled experiment conducted with a confederate inter- locutor. The interlocutor was asked to maintain either a soft, a modal or a loud voice level during the dialogues. Through half of the trials, subjects were exposed to a cocktail party noise through headphones. The analytical results suggest that both the background noise and the interlocutor’s voice level affect the dynamics of speaker entrainment. Speakers appear to still en- train to the voice level of their interlocutor in noisy conditions, though to a lesser extent, as strategies of ensuring intelligibility affect voice levels as well. These findings could be leveraged in spoken dialogue systems and speech generating devices to help choose a vocal effort level for the synthetic voice that is both intelligible and socially suited to a specific interaction.

Audiobooks are known to contain a variety of expressive speaking styles that occur as a result of the narrator mimicking a character in a story, or expressing affect. An accurate modeling of this variety is essential for the purposes of speech synthesis from an audiobook. Voice quality differences are important features characterizing these different speaking styles, which are realized on a gradient and are often difficult to predict from the text. The present study uses a parameter characterizing breathy to tense voice qualities using features of the wavelet transform, and a measure for identifying creaky segments in an utterance. Based on these features, a combination of supervised and unsupervised classification is used to detect the regions in an audiobook, where the speaker changes his regular voice quality to a particular voice style. The target voice style candidates are selected based on the agreement of the supervised classifier ensemble output, and evaluated in a listening test.

Freely available audiobooks are a rich resource of expressive speech recordings that can be used for the purposes of speech synthesis. Natural sounding, expressive synthetic voices have previously been built from audiobooks that contained large amounts of highly expressive speech recorded from a professionally trained speaker. The majority of freely available audiobooks, however, are read by amateur speakers, are shorter and contain less expressive (less emphatic, less emotional, etc.) speech both in terms of quality and quantity. Synthesizing expressive speech from a typical online audiobook therefore poses many challenges. In this work we address these challenges by applying a method consisting of minimally supervised techniques to align the text with the recorded speech, select groups of expressive speech segments and build expressive voices for hidden Markov-model based synthesis using speaker adaptation. Subjective listening tests have shown that the expressive synthetic speech generated with this method is often able to produce utterances suited to an emotional message. We used a restricted amount of speech data in our experiment, in order to show that the method is generally applicable to most typical audiobooks widely available online.

A great challenge for text-to-speech synthesis is to produce expressive speech. The main problem is that it is difficult to synthesise high-quality speech using expressive corpora. With the increasing interest in audiobook corpora for speech synthesis, there is a demand to synthesise speech which is rich in prosody, emotions and voice styles. In this work, Self-Organising Feature Maps (SOFM) are used for clustering the speech data using voice quality parameters of the glottal source, in order to map out the variety of voice styles in the corpus. Subjective evaluation showed that this clustering method successfully separated the speech data into groups of utterances associated with different voice characteristics. This work can be applied in unitselection synthesis by selecting appropriate data sets to synthesise utterances with specific voice styles. It can also be used in parametric speech synthesis to model different voice styles separately. Index Terms: expressive speech, voice quality, audiobook, speech synthesis