From 95b35a3f00579707e136d0de1b13b52e97037146 Mon Sep 17 00:00:00 2001
From: Aya <220561865+ayaadev@users.noreply.github.com>
Date: Fri, 27 Mar 2026 11:14:10 +0000
Subject: [PATCH 1/4] Fix outdated AudioSet link

Pin the link to a specific revision of the HuggingFace repository.
---
 setup-data.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/setup-data.sh b/setup-data.sh
index b5a5814..3274700 100755
--- a/setup-data.sh
+++ b/setup-data.sh
@@ -67,7 +67,7 @@ if [ ! -d "$DATA_DIR/audioset_16k" ]; then
     echo "Downloading AudioSet background audio..."
     mkdir -p "$DATA_DIR/audioset" "$DATA_DIR/audioset_16k"
     curl -L -o "$DATA_DIR/audioset/bal_train09.tar" \
-        'https://huggingface.co/datasets/agkphysics/AudioSet/resolve/main/bal_train09.tar'
+        'https://huggingface.co/datasets/agkphysics/AudioSet/resolve/728fffe80088808b49db5aa700f2e264c035c880/data/bal_train09.tar'
     tar -xf "$DATA_DIR/audioset/bal_train09.tar" -C "$DATA_DIR/audioset"
 
     python3 << EOF

From 3fb4f755951971f8bb4eecbb067f8fe3eabf6788 Mon Sep 17 00:00:00 2001
From: Aya <220561865+ayaadev@users.noreply.github.com>
Date: Sat, 28 Mar 2026 08:03:08 +0000
Subject: [PATCH 2/4] Set IPC to host in Docker compose

This fixes an issue where the container runs out of memory
---
 docker-compose.yml | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/docker-compose.yml b/docker-compose.yml
index 1bba6dc..a86fcc6 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -1,6 +1,7 @@
 services:
   kokoro:
     image: ghcr.io/remsky/kokoro-fastapi-gpu:latest
+    ipc: "host"
     ports:
       - "8880:8880"
     deploy:
@@ -13,6 +14,7 @@ services:
 
   trainer:
     build: .
+    ipc: "host"
     deploy:
       resources:
         reservations:

From 9986f5e7f71d90ca468cfa731c7683048961c030 Mon Sep 17 00:00:00 2001
From: Aya <220561865+ayaadev@users.noreply.github.com>
Date: Sat, 28 Mar 2026 08:04:30 +0000
Subject: [PATCH 3/4] Pin protobuf package to 3.20.x

This fixes the following error. Method one was used.

TypeError: Descriptors cannot be created directly.
If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.
If you cannot immediately regenerate your protos, some other possible workarounds are:
 1. Downgrade the protobuf package to 3.20.x or lower.
 2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).
---
 requirements.txt | 1 +
 1 file changed, 1 insertion(+)

diff --git a/requirements.txt b/requirements.txt
index 2b0f773..80ad6d7 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -20,6 +20,7 @@ acoustics==0.2.6
 pronouncing==0.2.0
 datasets==2.14.6
 deep-phonemizer==0.0.19
+protobuf>=3.20.0,<4.0.0
 
 # Data download/processing
 piper-phonemize

From 7eb86273e1e02d411646443cfa0d46efbb23cfd2 Mon Sep 17 00:00:00 2001
From: Aya <220561865+ayaadev@users.noreply.github.com>
Date: Sat, 28 Mar 2026 08:11:13 +0000
Subject: [PATCH 4/4] Inform the user that the finished model could be small

---
 README.md | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/README.md b/README.md
index 4ecfaf0..7455c44 100644
--- a/README.md
+++ b/README.md
@@ -56,6 +56,8 @@ docker compose run --rm trainer python train.py --wake-word "hey cal" --data-dir
 
 Training takes 4-8 hours depending on GPU.
 
+Don't be alarmed if the finished model is only a few hundred kilobytes. Test your model in the following section to see if it works correctly.
+
 ### 5. Test Your Model
 
 Test on your host machine (needs microphone access):