Add private realtime Talk for OpenClaw

platform/docker/whisper/entrypoint.sh

@@ -1,8 +1,8 @@
 #!/bin/sh
 set -eu
 
-model=${WHISPER_MODEL:-/models/ggml-large-v3-turbo.bin}
-model_url=${WHISPER_MODEL_URL:-https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo.bin}
+model=${WHISPER_MODEL:-/models/ggml-small.bin}
+model_url=${WHISPER_MODEL_URL:-https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.bin}
 model_dir=$(dirname "$model")
 
 mkdir -p "$model_dir"
@@ -22,4 +22,44 @@ if [ ! -s "$model" ]; then
   gosu whisper mv "$partial" "$model"
 fi
 
-exec gosu whisper python /app/stt_worker.py
+server_port=${WHISPER_SERVER_PORT:-8085}
+server_log=/tmp/whisper-server.log
+
+gosu whisper /opt/whisper.cpp/build/bin/whisper-server \
+  --model "$model" \
+  --threads "${WHISPER_THREADS:-8}" \
+  --language "${WHISPER_LANGUAGE:-de}" \
+  --host 127.0.0.1 \
+  --port "$server_port" \
+  --inference-path /inference \
+  --no-gpu >"$server_log" 2>&1 &
+
+server_pid=$!
+worker_pid=""
+cleanup() {
+  [ -z "$worker_pid" ] || kill "$worker_pid" 2>/dev/null || true
+  kill "$server_pid" 2>/dev/null || true
+}
+trap cleanup EXIT INT TERM
+
+# Der Modell-Load dauert beim Containerstart kurz. Der HTTP-Wrapper wird erst
+# freigegeben, wenn whisper-server tatsächlich Antworten annimmt.
+i=0
+until curl --fail --silent "http://127.0.0.1:${server_port}/" >/dev/null; do
+  i=$((i + 1))
+  if ! kill -0 "$server_pid" 2>/dev/null; then
+    cat "$server_log"
+    exit 1
+  fi
+  if [ "$i" -ge 180 ]; then
+    echo "whisper-server did not become ready" >&2
+    cat "$server_log"
+    exit 1
+  fi
+  sleep 1
+done
+
+export WHISPER_SERVER_URL="http://127.0.0.1:${server_port}"
+gosu whisper python /app/stt_worker.py &
+worker_pid=$!
+wait "$worker_pid"