<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="du-etal-2026-tracing">
    <titleInfo>
        <title>Tracing a Sparse Emotion-Control Circuit in LLM-Based Text-to-Speech</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Hongfei</namePart>
        <namePart type="family">Du</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jiacheng</namePart>
        <namePart type="family">Shi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yanfu</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ashley</namePart>
        <namePart type="family">Gao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-459-0</identifier>
    </relatedItem>
    <abstract>LLM-based text-to-speech (TTS) models can generate emotionally expressive speech, but how reference emotion is routed through the model and realized in decoded speech remains unclear. We introduce two emotion-sensitive metrics for matched neutral and emotional syntheses—a codec trajectory score and a late residual direction score—and use them to score activation-patching interventions. Under controlled matched-reference conditions, this analysis identifies a sparse source-to-readout component-level circuit: 23–27 attention heads and MLPs per emotion, roughly 5% of the components considered, recover or suppress 74–88% of the late emotion-readout shift on held-out cases. The circuit combines a shared component backbone with emotion-specific components; cross-emotion activation swaps reduce the target readout in 47 of 48 cases. In decoded speech, the same intervention produces consistent changes in pitch, energy, and spectral brightness over 24 matched pairs per emotion. A readout-matched residual-direction baseline produces only 17–27% of the intervention’s pitch effect, showing that internal readout movement alone does not explain the decoded acoustic changes. These results trace a compact causal route from reference-derived prefix information to emotion-relevant properties of generated speech.</abstract>
    <identifier type="citekey">du-etal-2026-tracing</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.378/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>8645</start>
            <end>8658</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Tracing a Sparse Emotion-Control Circuit in LLM-Based Text-to-Speech
%A Du, Hongfei
%A Shi, Jiacheng
%A Zhang, Yanfu
%A Gao, Ashley
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F du-etal-2026-tracing
%X LLM-based text-to-speech (TTS) models can generate emotionally expressive speech, but how reference emotion is routed through the model and realized in decoded speech remains unclear. We introduce two emotion-sensitive metrics for matched neutral and emotional syntheses—a codec trajectory score and a late residual direction score—and use them to score activation-patching interventions. Under controlled matched-reference conditions, this analysis identifies a sparse source-to-readout component-level circuit: 23–27 attention heads and MLPs per emotion, roughly 5% of the components considered, recover or suppress 74–88% of the late emotion-readout shift on held-out cases. The circuit combines a shared component backbone with emotion-specific components; cross-emotion activation swaps reduce the target readout in 47 of 48 cases. In decoded speech, the same intervention produces consistent changes in pitch, energy, and spectral brightness over 24 matched pairs per emotion. A readout-matched residual-direction baseline produces only 17–27% of the intervention’s pitch effect, showing that internal readout movement alone does not explain the decoded acoustic changes. These results trace a compact causal route from reference-derived prefix information to emotion-relevant properties of generated speech.
%U https://aclanthology.org/2026.emnlp-main.378/
%P 8645-8658

Markdown (Informal)

[Tracing a Sparse Emotion-Control Circuit in LLM-Based Text-to-Speech](https://aclanthology.org/2026.emnlp-main.378/) (Du et al., EMNLP 2026)

• Tracing a Sparse Emotion-Control Circuit in LLM-Based Text-to-Speech (Du et al., EMNLP 2026)

ACL

• Hongfei Du, Jiacheng Shi, Yanfu Zhang, and Ashley Gao. 2026. Tracing a Sparse Emotion-Control Circuit in LLM-Based Text-to-Speech. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 8645–8658, Budapest, Hungary. Association for Computational Linguistics.