updated readme with experiemtns

README.md

@@ -68,6 +68,26 @@ config_path = hf_hub_download(
 | Duck Audio | <audio controls src="https://huggingface.co/iliasslasri/robust_speech_quantizer/resolve/main/augmentations/13_duck_audio.wav"></audio> |
 | Up-Down Resample | <audio controls src="https://huggingface.co/iliasslasri/robust_speech_quantizer/resolve/main/augmentations/14_updownresample.wav"></audio> |
 
+## Experiments
+
+We trained quantizers across different encoders, codebook sizes, and augmentation strategies. The augmentation configurations are:
+
+- **All augmentations, chained** — all augmentations from the table above are enabled, and multiple augmentations are applied sequentially to each sample. The number of chained augmentations is sampled from a uniform distribution between 0 and 4.
+- **All augmentations, single** — all augmentations are enabled, but only one randomly chosen augmentation is applied per sample.
+- **No extra augmentations, single** — only the baseline augmentations (from the original paper) are used, with one applied per sample.
+
+| Encoder | Codebook | Augmentation Strategy |
+|:---|:---:|:---|
+| HuBERT | 500 | All augmentations, chained |
+| HuBERT | 500 | All augmentations, single |
+| HuBERT | 500 | No extra augmentations, single |
+| | | |
+| SpidR | 256 | No extra augmentations, single |
+| SpidR | 256 | All augmentations, chained |
+| | | |
+| DinoSR (original) | 256 | All augmentations, chained |
+| DinoSR (reproduced) | 256 | All augmentations, chained |
+
 ## Links
 - Paper: [Algayres et al., Interspeech 2023](https://aclanthology.org/2023.iwslt-1.46/)
 - Code: [GitHub](https://github.com/iliasslasri/snlp_project)