Speech Emotion Recognition
Recognising emotion, arousal and valence from the sound of speech across languages and recording conditions.
Tier Grades
Tier A: Frontier AI labs
feasible
Alibaba released emotion2vec as a universal emotion representation.
Tier B: Indonesian AI labs
feasible
Training only linear layers on emotion2vec, as the paper does, makes Indonesian adaptation cheap.
Tier C: Institution units such as ITB
feasible
A 7-author team built emotion2vec on open-source data.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Released models can flag caller emotion in contact-centre proofs of concept, subject to privacy rules.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeNot in the sources read: the emotion2vec abstract gives no compute; only linear layers are trained for the downstream task.
- Dataemotion2vec: pre-trained on open-source unlabelled emotion data, tested on IEMOCAP and 10 languages of emotion datasets.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 7 authors on emotion2vec.
- Time horizonMonths.
- Shelf-life signalsAudio language models reporting emotion as a standard output, Regulation limiting emotion recognition in workplaces
- Pipeline roleTier A and Tier C release emotion encoders; Tier B labels Indonesian emotional speech; Tier D applies them in call analytics pilots.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Links
R&D TopicSpeech Emotion Recognition
Models
emotion2vec+ largeAlibaba (Qwen), audio classification
wav2vec2-large-robust-12-ft-emotion-msp-dimaudEERING, audio classification
Emotion recognition wav2vec2 (IEMOCAP)SpeechBrain community, audio classificationSenseVoiceSmallAlibaba (Qwen), speech recognition+1 more
emotion2vec+ largeAlibaba (Qwen), audio classificationwav2vec2-large-robust-12-ft-emotion-msp-dimaudEERING, audio classificationEmotion recognition wav2vec2 (IEMOCAP)SpeechBrain community, audio classificationSenseVoiceSmallAlibaba (Qwen), speech recognition+1 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 5
SenseVoiceSmallAlibaba (Qwen), speech recognitionproduced by
TRILLssonGoogle, embeddingused byEmotion recognition wav2vec2 (IEMOCAP)SpeechBrain community, audio classificationproduced byemotion2vec+ largeAlibaba (Qwen), audio classificationexemplarwav2vec2-large-robust-12-ft-emotion-msp-dimaudEERING, audio classificationexemplarPapers
- SUPERB: Speech processing Universal PERformance BenchmarkShu-wen Yang et al., Interspeech 2021 (per arXiv comment), 2021 founding
- EmoBox: Multilingual Multi-corpus Speech Emotion Recognition Toolkit and BenchmarkZiyang Ma et al., INTERSPEECH 2024 (per arXiv comment), 2024 state of the art
- emotion2vec: Self-Supervised Pre-Training for Speech Emotion RepresentationZiyang Ma et al., arXiv preprint, 2023 state of the art
- Emotion Recognition from Speech Using Wav2vec 2.0 EmbeddingsLeonardo Pepino et al., Interspeech 2021 (per arXiv comment), 2021 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2312.15185 https://arxiv.org/abs/2312.15185
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Speech Emotion Recognition", https://aurion.semai.id/topics/speech-emotion-recognition/, records of October 2026.