Bibliographic record
Abstract
Biometric identification technology has been widely used in today's society because of its convenience and security. As an important biometric feature, speech contains abundant information, and because of the popularity of smart devices, the collection cost of a speaker's speech is also very low. Therefore, it is of great practical value to analyze the speaker's voice. This paper mainly discusses the speaker's voiceprint recognition based on deep learning and expands the speech emotion recognition. Voiceprint recognition is divided into two tasks: speaker identification and speaker confirmation, and speech emotion recognition will be directly treated as a multi-classification problem. To take advantage of different attention mechanisms, this paper proposes a dual-path attention mechanism, which applies self-attention and convolution module attention at the same time and significantly improves the recognition effect without increasing the training time. Based on the ternary loss of predecessors, the cluster domain loss is proposed, and this paper further improves this loss for the speaker identification task and puts forward the weighted cluster domain loss, which pays more attention to the increase of the difference between classes, thus increasing the probability that critical samples are correctly classified. To solve the problem of low efficiency of cluster loss in the early stage of training, this paper also puts forward a novel loss function-critical enhancement loss, which pays extra attention to the sample pairs that are easiest and necessary to be optimized in every step of the training process. After combining cluster loss, the samples that are most difficult to optimize and easiest to optimize in every step are considered at the same time, which accelerates the training process and wins more training time for the difficult loss in cluster loss, thus further improving the final optimization effect. Aiming at the task of speaker emotion recognition, this paper proposes a lightweight neural network that combines Res Net and GRU. Compared with other methods in newer literature, this paper achieves comparable emotion classification results on the IEMOCAP data set with fewer parameters and features, in which UA reaches 67.9%, F1 score reaches 0.675, and the number of parameters is relatively reduced by 16.2%.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.016 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".