From b75a11478c2f68230d2a956fb68ab612c4d2c34f Mon Sep 17 00:00:00 2001
From: Nay San <nay.san@stanford.edu>
Date: Tue, 14 Sep 2021 19:18:23 -0700
Subject: [PATCH] Forgot to stage one file

---
 scripts/README.md | 56 ++++++++++++++++++++++++++++-------------------
 1 file changed, 34 insertions(+), 22 deletions(-)

diff --git a/scripts/README.md b/scripts/README.md
index 5b3ef32..1643fe6 100644
--- a/scripts/README.md
+++ b/scripts/README.md
@@ -70,21 +70,7 @@ mkdir -p data/raw/datasets/gos-kdl
 unzip tmp/gos-kdl.zip -d data/raw/datasets/gos-kdl
 ``` 
 
-### 1.3 Fetch model checkpoint(s)
-
-The exact model checkpoint files used in our pilot and main experiments have been placed on Zenodo ([https://zenodo.org/record/4632537](https://zenodo.org/record/4632537)).
-
-```bash
-# Fetch the Librispeech 960 checkpoint
-wget https://zenodo.org/record/4632537/files/20210225-Large-0FT.pt \
-   -P data/raw/model_checkpoints/
-   
-# Fetch the XLSR-53 checkpoint (optional)
-wget https://zenodo.org/record/4632537/files/20210127-XLSR53.pt \
-   -P data/raw/model_checkpoints/
-```
-
-#### 1.4 Pull docker image(s)
+#### 1.3 Pull docker image(s)
 
 ```bash
 # For extracting wav2vec 2.0 features and running evaluation scripts
@@ -130,6 +116,30 @@ exit
 ```
 
 ### 2.2 wav2vec 2.0 features
+
+We use Hugging Face to help fetch the wav2vec 2.0 models to use for feature extraction. The model repo paths (e.g. `facebook/wav2vec2-base`) can be found in the `wav_to_w2v2-feats.py` script (note: for reproducibility of the analyses, the `wav2vec2-large` and `wav2vec2-large-xlsr-53` have specific model versions):
+
+```python
+KNOWN_MODELS = {
+    # Pre-trained
+    'wav2vec2-base': 'facebook/wav2vec2-base',
+    'wav2vec2-large': {'name' : 'facebook/wav2vec2-large', 'revision' : '85c73b1a7c1ee154fd7b06634ca7f42321db94db' },
+    # March 11, 2021 version: https://huggingface.co/facebook/wav2vec2-large/commit/85c73b1a7c1ee154fd7b06634ca7f42321db94db
+    'wav2vec2-large-lv60': 'facebook/wav2vec2-large-lv60',
+    'wav2vec2-large-xlsr-53': {'name' : 'facebook/wav2vec2-large-xlsr-53', 'revision' : '8e86806e53a4df405405f5c854682c785ae271da' },
+    # May 6, 2021 version: https://huggingface.co/facebook/wav2vec2-large-xlsr-53/commit/8e86806e53a4df405405f5c854682c785ae271da
+    
+    # Fine-tuned
+    'wav2vec2-base-960h': 'facebook/wav2vec2-base-960h',
+    'wav2vec2-large-960h': 'facebook/wav2vec2-large-960h',
+    'wav2vec2-large-960h-lv60': 'facebook/wav2vec2-large-960h-lv60',
+    'wav2vec2-large-960h-lv60-self': 'facebook/wav2vec2-large-960h-lv60-self',
+    'wav2vec2-large-xlsr-53-english': 'jonatasgrosman/wav2vec2-large-xlsr-53-english',
+    'wav2vec2-large-xlsr-53-tamil': 'manandey/wav2vec2-large-xlsr-tamil'
+}
+```
+
+To extract features using one of these models run:
 
 ```bash
 # Start docker container according to 'dev' config
@@ -137,16 +147,18 @@ exit
 docker-compose run --rm dev
 
 # Extract features from all stages/layers (encoder, quantizer, transformer 1-24)
-# of wav2vec 2.0 model using model weights from specified checkpoint file.
+# of wav2vec 2.0 model (wav2vec2-large, revision: 85c73b)
 #
 # For help, run: python scripts/wav_to_w2v2-feats.py -h
+
+python scripts/wav_to_w2v2-xlsr-feats.py \
+   --dataset gos-kdl \
+   --stage _all_ \
+   --layer _all_ \
+   --model wav2vec2-large
+```
+
 
-python scripts/wav_to_w2v2-feats.py \
-    data/raw/model_checkpoints/20210225-Large-0FT.pt \
-    gos-kdl \
-    --stage _all_ \
-    --layer _all_
-```
 
 ### 2.3 Fetch features from Zenodo (optional)