Analysis of emotion recognition using facial expressions, speech and multimodal information
The DOI matches its Crossref record; 24 of the 26 references it cites were found in a registry or an index; 4 of the 15 sources its claims cite were checked for support.
Abstract
The interaction between human beings and computers will be more natural if computers are able to perceive and respond to human non-verbal communication such as emotions. Although several approaches have been proposed to recognize human emotions based on facial expressions or speech, relatively limited work has been done to fuse these two, and other, modalities to improve the accuracy and robustness of the emotion recognition system. This paper analyzes the strengths and the limitations of systems based only on facial expressions or acoustic information. It also discusses two approaches used to fuse these two modalities: decision level and feature level integration. Using a database recorded from an actress, four emotions were classified: sadness, anger, happiness, and neutral state. By the use of markers on her face, detailed facial motions were captured with motion capture, in conjunction with simultaneous speech recordings. The results reveal that the system based on facial expression gave better performance than the system based on just acoustic information for the emotions considered. Results also show the complementarily of the two modalities and that when these two modalities are fused, the performance and the robustness of the emotion recognition system improve measurably.
Overview
CoChat Summary
TL;DR
This paper compares emotion recognition from speech, facial expressions, and fused audio-visual information for sadness, anger, happiness, and neutral state. It finds that facial-expression recognition outperformed acoustic recognition, and that fusing the two modalities improved overall performance, with feature-level and decision-level fusion giving similar overall accuracy but different per-emotion strengths.
Key findings
- The acoustic classifier achieved 70.9% overall accuracy; sadness was misclassified as neutral 22% of the time and neutral as sadness 14%, while happiness and anger were mutually confused (19% and 21%, respectively).
- The combined facial-expression classifier achieved 85.1% overall accuracy, with happiness recognized at 1.00, sadness at 0.81, anger at 0.79, and neutral at 0.81; cheek areas gave valuable information, while eyebrows gave the poorest performance among facial blocks.
- Feature-level bimodal fusion achieved 89.1% overall accuracy, with anger 0.95, sadness 0.79, happiness 0.91, and neutral 0.92; sadness remained confused with neutral (18%).
- Decision-level bimodal fusion with product combining achieved 89.0% overall accuracy, with anger 0.84, sadness 0.90, happiness 0.98, and neutral 0.84; the paper reports that bimodal fusion improved almost 5 percentage points absolute over the facial-expression system.
Methods
- Data came from one actress who read 258 sentences expressing four emotions; facial motion was captured with a VICON system using three cameras at 120 Hz and 102 facial markers, with simultaneous speech recorded at 48 kHz.
- Three systems used support vector machine classifiers with 2nd-order polynomial kernels and leave-one-out cross validation: an acoustic system using 11-dimensional prosodic/voiced-unvoiced features, a facial-expression system using PCA-reduced marker data from five facial blocks, and bimodal systems using feature-level or decision-level fusion.
- Facial features were reduced to 10-dimensional vectors per block via PCA covering more than 99% of variation, classified at frame level with K-nearest neighbor (k=3), then aggregated into 4-dimensional utterance-level vectors per block; bimodal feature-level fusion selected 10 features, while decision-level fusion combined posterior probabilities using maximum, average, product, or weight criteria.
Limitations
- The paper states that the database was recorded from a single actress, so more data from other people are needed to represent human variability; it also states that facial markers are not feasible for real applications and that automatic markerless facial-motion extraction should be implemented.
- The paper states that only four affective states were studied, so eyebrows may still matter for other emotions such as surprise.
- Evident from the text: the emotional speech was acted by one speaker, and the reported cross-validation is within that single-speaker database, so cross-subject generalization is not established by these experiments.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
3 Valid · 1 Not valid · 5 Not checked · 79 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“However, none of these papers attempted to compare which fusion approach is more suitable for emotion recognition.”
What the sources say
- Disputedcites [4] [4] Chen, L.S., Huang, T. S., Miyasato T., and Nakatsu R. Multimodal human emotion / expression recognition, in Proc. of Int. Conf. on Automatic Face and Gesture Recognition, (Nara, Japan), IEEE Computer Soc., April 1998 sourceExperimental — accuracy not yet measured
Section 5.2 (p. 4) assesses the difficulty of preliminary classification in the existing audio/video approach and says a different way of combining the modalities is necessary. Section 6 (p. 5) then proposes a rule-based alternative. This is a qualitative assessment of which approach is suitable, though the paper does not report a head-to-head performance comparison. (LLM, conf=0.76)
Checked by a reasoning model — may be wrong.
- Not checked — no full text of the sourcecites [8] [8] De Silva, L.C., Ng, P. C. Bimodal emotion recognition. Automatic Face and Gesture Recognition, 2000. Proceedings. Fourth IEEE International Conference on, 28- 30 March 2000. Pages: 332 – 335. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Although several automatic emotion recognition systems have explored the use of either facial expressions [1],[11],[16],[21],[22] or speech [9],[18],[14] to detect human affective states, relatively few efforts have focused on emotion recognition using both modalities [4],[8].”
- Also stated in: § 2.3 Emotion recognition by bimodal data · plain
“Relatively few efforts have focused on implementing emotion recognition systems using both facial expressions and acoustic information.”
What the sources say
- Supportedcites [4] [4] Chen, L.S., Huang, T. S., Miyasato T., and Nakatsu R. Multimodal human emotion / expression recognition, in Proc. of Int. Conf. on Automatic Face and Gesture Recognition, (Nara, Japan), IEEE Computer Soc., April 1998 source
The Introduction (section 1) says most emotion-recognition studies used speech or video alone, while relatively little work combined the two. The Abstract makes the same point about combining facial images and speech to recognize emotion. (LLM, conf=0.95)
- Not checked — no full text of the sourcecites [8] [8] De Silva, L.C., Ng, P. C. Bimodal emotion recognition. Automatic Face and Gesture Recognition, 2000. Proceedings. Fourth IEEE International Conference on, 28- 30 March 2000. Pages: 332 – 335. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 2.3 Emotion recognition by bimodal data · plain
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“All these studies have shown that the performance of emotion recognition systems can be improved by the use of multimodal information.”
§ 2. EMOTION RECOGNITION SYSTEMS > 2.3 Emotion recognition by bimodal data
What the sources say
- Supportedcites [4] [4] Chen, L.S., Huang, T. S., Miyasato T., and Nakatsu R. Multimodal human emotion / expression recognition, in Proc. of Int. Conf. on Automatic Face and Gesture Recognition, (Nara, Japan), IEEE Computer Soc., April 1998 source
The Abstract states that combining audio and video can achieve higher emotion-recognition rates than either modality alone. The Conclusions likewise describe a bimodal algorithm that can give a higher recognition rate. (LLM, conf=0.94)
- Not checked — no full text of the sourcecites [8] [8] De Silva, L.C., Ng, P. C. Bimodal emotion recognition. Automatic Face and Gesture Recognition, 2000. Proceedings. Fourth IEEE International Conference on, 28- 30 March 2000. Pages: 332 – 335. source
- Not matchedcites [12] [12] Huang, T. S., Chen, L. S., Tao, H., Miyasato, T., Nakatsu, R. Bimodal Emotion Recognition by Man and Machine. Proceeding of ATR Workshop on Virtual Communication Environments, (Kyoto, Japan), April 1998.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“These results agree with the finding reported by Chen et al. [4], which showed that audio and facial expressions data present complementary information.”
What the sources say
- Supportedcites Chen et al. [4] [4] Chen, L.S., Huang, T. S., Miyasato T., and Nakatsu R. Multimodal human emotion / expression recognition, in Proc. of Int. Conf. on Automatic Face and Gesture Recognition, (Nara, Japan), IEEE Computer Soc., April 1998 source
The Abstract explicitly says the authors found audio and video modalities complementary and achieved higher recognition rates using both than using either alone. The Conclusions (p. 5) also describe a recognition algorithm based on that complementary property. (LLM, conf=0.94)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The diagonal components of table 1 reveal that all the emotions can be recognized with more than 64 percent of accuracy, by using only the features of the speech.”
§ 4. RESULTS > 4.1 Acoustic emotion classifier
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
55 more claims
Create a free account to see all 88 claims and every source verdict.
- It may state two separate claims as one. 19
- The quote only partly states this claim. 1
- Probably not a substantial claim. 11
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“Weight criterion solves this by assigning a confidence to each classifier output, otputting a weighted linear combination of the predictions [161], [162], [167], [200].”
From Survey on RGB, 3D, Thermal, and Multimodal Approaches for Facial Expression Recognition: History, Trends, and Affect-Related Applications (2016) · cites this paper as [162]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4. RESULTS > 4.2 System based on facial expressions
“Table 3 shows the performance of the emotion recognition systems based on facial expressions, for each of the five facial blocks and the combined facial expression classifier.”
§ 4. RESULTS > 4.2 System based on facial expressions
“Table 2: Performance of the facial expression classifiers”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4. RESULTS > 4.3 Bimodal system
“These results suggest that in the decision-level fusion approach, the recognition rate of each emotion is increased, improving the performance of the bimodal system.”
§ 5. DISCUSSION
“In the decision- level bimodal classifier, the recognition rate of each emotion increased compared to the facial expression classifier, which was the best unimodal recognition system (except happiness, which decreased in 2%).”
Citations
Cites
A Tutorial on Support Vector Machines for Pattern Recognition
Burges, C. J. C. (1998) · Data Mining and Knowledge Discovery · doi:10.1023/a:1009715923555 · Cited by 16,429
Ekman, P., & Friesen, W. V. (2019) · PsycTESTS Dataset · doi:10.1037/t27734-000 · Cited by 6,019
A TUTORIAL ON SUPPORT VECTOR MACHINES FOR PATTERN RECOGNITION, DATA MINING AND KNOWLEDGE DISCOVERY
Burger, C. J. C. (1998) · Cited by 3,313
Emotion recognition in human-computer interaction
Cowie, R., Douglas-Cowie, E., Tsapatsoulis, N., Votsis, G., Kollias, S., Fellenz, W., & Taylor, J. G. (2001) · IEEE Signal Processing Magazine · doi:10.1109/79.911197 · Cited by 2,603
Toward detecting emotions in spoken dialogs
Lee, C. M., & Narayanan, S. S. (2005) · IEEE Transactions on Speech and Audio Processing · doi:10.1109/tsa.2004.838534 · Cited by 988
Self-citationCoding, analysis, interpretation, and recognition of facial expressions
Essa, I. A., & Pentland, A. P. (1997) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/34.598232 · Cited by 911
Toward an affect-sensitive multimodal human-computer interaction
Pantic, M., & Rothkrantz, L. J. M. (2003) · Proceedings of the IEEE · doi:10.1109/jproc.2003.817122 · Cited by 791
Recognition of Facial Expression from Optical Flow
Mase, K. (1991) · IEICE Transactions on Information and Systems · Cited by 480
Black, M. J., & Yacoob, Y. (2002) · Proceedings of IEEE International Conference on Computer Vision · doi:10.1109/iccv.1995.466915 · Cited by 457
Facial emotion recognition using multi-modal information
De Silva, L. C., Miyasato, T., & Nakatsu, R. (2002) · Proceedings of ICICS, 1997 International Conference on Information, Communications and Signal Processing. Theme: Trends in Information Systems Engineering and Wireless Multimedia Communications (Cat. No.97TH8237) · doi:10.1109/icics.1997.647126 · Cited by 319
Research works printed but not linked: 7
- 10.5555/839277.840053 · doi:10.5555/839277.840053
- Boersma , P. , Weenink , D. , Praat Speech Processing Software , Institute of Phonetics Sciences of the University of Amsterdam. http ://www.praat.org Boersma, P., Weenink, D., Praat Speech Processing Software, Institute of Phonetics Sciences of the University of Amsterdam. http://www.praat.org
- 10.5555/520809.796063 · doi:10.5555/520809.796063
- De Silva , L. C. , Miyasato , T. , and Nakatsu , R . Facial Emotion Recognition Using Multimodal Information . In Proc. IEEE Int. Conf. on Information, Communications and Signal Processing (ICICS'97) , Singapore , pp. 397 -- 401 , Sept. 1997 . De Silva, L. C., Miyasato, T., and Nakatsu, R. Facial Emotion Recognition Using Multimodal Information. In Proc. IEEE Int. Conf. on Information, Communications and Signal Processing (ICICS'97), Singapore, pp. 397--401, Sept. 1997.
- 10.5555/795661.796218 · doi:10.5555/795661.796218
- 10.5555/524467.796011 · doi:10.5555/524467.796011
- Yildirim , S. , Bulut , M. , Lee , C. M. , Kazemzadeh , A. , Deng , Z. , Busso , C. , Lee , S. , Narayanan , S.S. , Analysis of acoustic correlates in emotional speech. to appear in ICSLP'04 , 2004 . Yildirim, S., Bulut, M., Lee, C. M., Kazemzadeh, A., Deng, Z., Busso, C., Lee, S., Narayanan, S.S., Analysis of acoustic correlates in emotional speech. to appear in ICSLP'04, 2004.
Cited by
IEMOCAP: Interactive emotional dyadic motion capture database
Busso, C., Bulut, M., Lee, C.-C., Kazemzadeh, A., Mower, E., Kim, S., Chang, J. N., Lee, S., & Narayanan, S. S. (2008) · Language Resources and Evaluation · doi:10.1007/s10579-008-9076-6 · Cited by 3,869
Self-citationA Survey of Affect Recognition Methods: Audio, Visual, and Spontaneous Expressions
Zeng, Z., Pantic, M., Roisman, G. I., & Huang, T. S. (2009) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2008.52 · Cited by 2,797
Affect Detection: An Interdisciplinary Review of Models, Methods, and Their Applications
Calvo, R. A., & D'Mello, S. (2010) · IEEE Transactions on Affective Computing · doi:10.1109/t-affc.2010.1 · Cited by 1,893
A review of affective computing: From unimodal analysis to multimodal fusion
Poria, S., Cambria, E., Bajpai, R., & Hussain, A. (2017) · Information Fusion · doi:10.1016/j.inffus.2017.02.003 · Cited by 1,644
CREMA-D: Crowd-Sourced Emotional Multimodal Actors Dataset
Cao, H., Cooper, D. G., Keutmann, M. K., Gur, R. C., Nenkova, A., & Verma, R. (2014) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2014.2336244 · Cited by 914
“[52] found, the recognition rate at each intensity level is worst for audio-only and best for audio-visual.”
Background
Metadata
- Authors
- Busso, C., Deng, Z., Yildirim, S., Bulut, M., Lee, C. M., Kazemzadeh, A., Lee, S., Neumann, U., & Narayanan, S.
- Affiliations
- University of Southern California (US)
- University of Southern California, Los Angeles
- ORCID
- Venue
- Proceedings of the 6th international conference on Multimodal interfaces
- Publisher
- ACM
- Published
- 2004-10-13
- Pages
- 205-211
- OpenAlex
- W2143350951
- Semantic Scholar
- 069443f2bbeb2deedefb600c82ed59cde2137e60
- Cited by
- 840
- References
- 26
- Citations by year
- 2026: 15 · 2025: 25 · 2024: 38 · 2023: 33 · 2022: 36 · 2021: 50 · 2020: 64 · 2019: 63 · 2018: 66 · 2017: 56 · 2016: 70 · 2015: 39 · 2014: 35 · 2013: 41 · 2012: 32
- FWCI
- 27.41
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- INTERNATIONAL CONFERENCE ON MULTIMODAL INTERACTION (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face and Expression Recognition
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Keywords
- emotion recognition, facial expressions, speech, multimodal fusion, decision fusion, feature-level fusion, nonverbal communication, motion capture, sadness, anger, happiness, human-computer interaction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Analysis of emotion recognition using facial expressions, speech and multimodal information. https://app.cochat.ai/papers/10.1145/1027933.1027968
- The paper
- Busso, C., Deng, Z., Yildirim, S., Bulut, M., Lee, C. M., Kazemzadeh, A., Lee, S., Neumann, U., & Narayanan, S. (2004). Analysis of emotion recognition using facial expressions, speech and multimodal information. Proceedings of the 6th international conference on Multimodal interfaces. https://doi.org/10.1145/1027933.1027968
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.