Speech

• Analyze animal vocalization recognition results represented using the Inter-Species Phonetic Alphabet (ISPA)
• Evaluate animal-species classification performance using ISPA-based recognition results
• Cluster ISPA symbols using topic models
• Next step: conduct a more detailed evaluation using more data

• Automatically remove fillers such as “um” and “uh” during speech recognition
• Fine-tune the large-scale Open Whisper-style Speech Model (OWSM)
• Teach the model not to produce fillers in its recognition output
• Next step: extend the approach beyond filler removal to broader transcript editing

• Partially control pitch during text-to-speech synthesis
• Add symbols to the input text to specify the desired pitch
• Learn the relationship between the symbols and pitch using the PJS Japanese singing-voice corpus
• Next steps: expand the controllable pitch range and incorporate musical constraints other than pitch