Add EasyCom prep, Qwen2.5-Omni feature extraction, streaming turn model prototype, and docs for the ICASSP 2027 ego-speak study. Co-authored-by: Cursor <cursoragent@cursor.com>
10 lines
133 B
Plaintext
10 lines
133 B
Plaintext
torch
|
|
numpy
|
|
transformers==4.52.3
|
|
accelerate
|
|
qwen-omni-utils[decord]
|
|
librosa
|
|
Pillow
|
|
|
|
# Optional on compatible NVIDIA GPUs: flash-attn
|