Update TTS component

2025-08-21 21:53:39 +02:00
parent 576324fada
commit d80d3c545c
8 changed files with 82 additions and 122 deletions
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -20,15 +20,28 @@ services:
    networks:
      - nointernet

-  openedai-speech:
+  ollama_provision:
    build:
-      dockerfile: src/tts/Dockerfile
+      dockerfile: src/ollama_provision/Dockerfile
    environment:
-      - TTS_HOME=voices
+      - MODELS=nomic-embed-text,llama3.1:8b,qwen2.5:7b,gemma3:12b-it-qat,gemma3:27b-it-qat,gpt-oss:20b
    volumes:
-      - voices:/app/voices
-      - speech-config:/app/config
+      - ollama:/root/.ollama
+    restart: no
+    depends_on:
+      ollama:
+        condition: service_healthy
+        restart: true
+    links:
+      - ollama
+    networks:
+      - internet
+
+  speaches-ai:
+    image: ghcr.io/speaches-ai/speaches:latest-cuda
    restart: unless-stopped
+    volumes:
+      - hf-hub-cache:/home/ubuntu/.cache/huggingface/hub
    deploy:
      resources:
        reservations:
@@ -45,20 +58,19 @@ services:
    networks:
      - nointernet

-  ollama_provision:
+  speaches-ai_provision:
    build:
-      dockerfile: src/ollama_provision/Dockerfile
-    environment:
-      - MODELS=qwen2.5:latest,qwen2.5-coder:32b,nomic-embed-text:latest,gemma2:latest,mistral:latest,deepseek-r1:7b,deepcoder:14b
+      dockerfile: src/speaches-ai_provision/Dockerfile
    volumes:
-      - ollama:/root/.ollama
+      - hf-hub-cache:/home/ubuntu/.cache/huggingface/hub
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [gpu]
    restart: no
-    depends_on:
-      ollama:
-        condition: service_healthy
-        restart: true
-    links:
-      - ollama
    networks:
      - internet

@@ -89,17 +101,23 @@ services:
      - NGINX_ENVSUBST_OUTPUT_DIR=/etc/nginx
      - API_KEY=${LLM_API_KEY}
    depends_on:
-      - openedai-speech
      - ollama
+      - speaches-ai
      - aichat
    links:
      - ollama
-      - openedai-speech
+      - speaches-ai
      - aichat
    ports:
-      - "127.0.0.1:11434:11434"
+      - "127.0.0.1:10434:10434"
      - "127.0.0.1:8000:8000"
      - "127.0.0.1:8001:8001"
+    healthcheck:
+      test: curl --fail http://localhost:11434 || exit 1
+      interval: 10s
+      retries: 5
+      start_period: 10s
+      timeout: 10s
    restart: unless-stopped
    networks:
      - internet
@@ -107,8 +125,7 @@ services:

 volumes:
  ollama:
-  voices:
-  speech-config:
+  hf-hub-cache:

 networks:
  internet:
--- a/src/nginx/nginx.conf
+++ b/src/nginx/nginx.conf
@@ -30,7 +30,7 @@ http {
            return 403;
        }
        location / {
-            proxy_pass http://openedai-speech:8000;
+            proxy_pass http://speaches-ai:8000;
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
@@ -48,6 +48,9 @@ http {
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header X-Forwarded-Proto $scheme;
+	    proxy_http_version 1.1;
+	    proxy_set_header Upgrade $http_upgrade;
+	    proxy_set_header Connection "upgrade";
        }
    }
 }
--- a/src/speaches-ai_provision/Dockerfile
+++ b/src/speaches-ai_provision/Dockerfile
@@ -0,0 +1,9 @@
+FROM ghcr.io/speaches-ai/speaches:latest-cuda
+
+USER root
+ADD ./src/speaches-ai_provision/entrypoint.sh /entrypoint.sh
+RUN chmod 755 /entrypoint.sh
+
+USER ubuntu
+
+ENTRYPOINT ["/entrypoint.sh"]
--- a/src/speaches-ai_provision/entrypoint.sh
+++ b/src/speaches-ai_provision/entrypoint.sh
@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+
+pkill -f uvicorn
+source .venv/bin/activate
+uvicorn --factory speaches.main:create_app &
+sleep 5
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/Kokoro-82M-v1.0-ONNX-int8'
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/piper-fr_FR-tom-medium'
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/piper-fr_FR-upmc-medium'
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/piper-en_GB-alba-medium'
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/piper-en_GB-northern_english_male-medium'
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/piper-en_US-john-medium'
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/piper-en_US-bryce-medium'
+curl -s -X POST -d '' 'http://localhost:8000/v1/models/speaches-ai/piper-en_US-ryan-high'
--- a/src/tts/Dockerfile
+++ b/src/tts/Dockerfile
@@ -1,49 +0,0 @@
-FROM python:3.11-slim
-
-RUN --mount=type=cache,target=/root/.cache/pip pip install -U pip
-
-ARG TARGETPLATFORM
-RUN <<EOF
-apt-get update
-apt-get install --no-install-recommends -y curl ffmpeg git
-if [ "$TARGETPLATFORM" != "linux/amd64" ]; then
-	apt-get install --no-install-recommends -y build-essential
-	curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
-fi
-
-# for deepspeed support - image +7.5GB, over the 10GB ghcr.io limit, and no noticable gain in speed or VRAM usage?
-#curl -O https://developer.download.nvidia.com/compute/cuda/repos/debian11/x86_64/cuda-keyring_1.1-1_all.deb
-#dpkg -i cuda-keyring_1.1-1_all.deb
-#rm cuda-keyring_1.1-1_all.deb
-#apt-get install --no-install-recommends -y libaio-dev build-essential cuda-toolkit
-
-apt-get clean
-rm -rf /var/lib/apt/lists/*
-EOF
-#ENV CUDA_HOME=/usr/local/cuda
-ENV PATH="/root/.cargo/bin:${PATH}"
-
-WORKDIR /app
-RUN mkdir -p voices config
-
-ARG USE_ROCM
-ENV USE_ROCM=${USE_ROCM}
-
-RUN git clone https://github.com/matatonic/openedai-speech.git /tmp/app
-RUN mv /tmp/app/* /app/
-ADD src/tts/download_voices_tts-1.sh /app/download_voices_tts-1.sh
-ADD src/tts/voice_to_speaker.default.yaml /app/voice_to_speaker.default.yaml
-RUN if [ "${USE_ROCM}" = "1" ]; then mv /app/requirements-rocm.txt /app/requirements.txt; fi
-RUN --mount=type=cache,target=/root/.cache/pip pip install -r requirements.txt
-
-
-ARG PRELOAD_MODEL
-ENV PRELOAD_MODEL=${PRELOAD_MODEL}
-ENV TTS_HOME=voices
-ENV HF_HOME=voices
-ENV COQUI_TOS_AGREED=1
-
-RUN bash /app/download_voices_tts-1.sh
-
-CMD bash startup.sh
-
--- a/src/tts/download_voices_tts-1.sh
+++ b/src/tts/download_voices_tts-1.sh
@@ -1,8 +0,0 @@
-#!/bin/sh
-# cat voice_to_speaker.default.yaml | yq '.tts-1 ' | grep mode | cut -d'/' -f2 | cut -d'.' -f1 | sort -u | xargs
-models=${*:-"en_GB-alba-medium en_GB-northern_english_male-medium en_US-bryce-medium en_US-john-medium en_US-libritts_r-medium en_US-ryan-high fr_FR-siwis-medium fr_FR-tom-medium fr_FR-upmc-medium"}
-piper --update-voices --data-dir voices --download-dir voices --model x 2> /dev/null
-for i in $models ; do
-    [ ! -e "voices/$i.onnx" ] && piper --data-dir voices --download-dir voices --model $i < /dev/null > /dev/null
-done
-
--- a/src/tts/voice_to_speaker.default.yaml
+++ b/src/tts/voice_to_speaker.default.yaml
@@ -1,36 +0,0 @@
-# Use https://rhasspy.github.io/piper-samples/ to configure
-tts-1:
-  alloy:
-    model: voices/en_US-libritts_r-medium.onnx
-    speaker: 79 
-  siwis:
-    model: voices/fr_FR-siwis-medium.onnx
-    speaker: 0
-  tom:
-    model: voices/fr_FR-tom-medium.onnx
-    speaker: 0
-  pierre:
-    model: voices/fr_FR-upmc-medium.onnx
-    speaker: 1
-  jessica:
-    model: voices/fr_FR-upmc-medium.onnx
-    speaker: 0
-  alba:
-    model: voices/en_GB-alba-medium.onnx
-    speaker: 0
-  jack:
-    model: voices/en_GB-northern_english_male-medium.onnx
-    speaker: 0
-  john:
-    model: voices/en_US-john-medium.onnx
-    speaker: 0
-  bryce:
-    model: voices/en_US-bryce-medium.onnx
-    speaker: 0
-  ryan:
-    model: voices/en_US-ryan-high.onnx
-    speaker: 0
-  echo:
-    model: voices/en_US-libritts_r-medium.onnx
-    speaker: 134 
-
--- a/tools/speech.sh
+++ b/tools/speech.sh
@@ -25,21 +25,30 @@ fi

 tts_host=${TTS_API_HOST:-"http://localhost:8000"}

+declare -A voices_model=(
+    [ff_siwis]="speaches-ai/Kokoro-82M-v1.0-ONNX-int8"
+    [tom]="speaches-ai/piper-fr_FR-tom-medium"
+    [upmc]="speaches-ai/piper-fr_FR-upmc-medium"
+    [alba]="speaches-ai/piper-en_GB-alba-medium"
+    [northern_english_male]="speaches-ai/piper-en_GB-northern_english_male-medium"
+    [john]="speaches-ai/piper-en_US-john-medium"
+    [bryce]="speaches-ai/piper-en_US-bryce-medium"
+    [ryan]="speaches-ai/piper-en_US-ryan-high"
+)
+
 _choice_voice() {

 	if [[ "${argc_lang}" == "fr" ]]; then
-		echo siwis
+		echo ff_siwis
 		echo tom
-		echo pierre
-		echo jessica
+		echo upmc
 	fi
 	if [[ "${argc_lang}" == "en" ]]; then
 		echo alba
-		echo jack
+		echo northern_english_male
 		echo john
 		echo bryce
 		echo ryan
-		echo echo
 	fi
 }

@@ -52,10 +61,11 @@ _choice_voice() {
 # @arg         text!                                    Set the text
 synthesize() {

-	http_status_code=$(curl -s "${tts_host}/v1/audio/speech" -o "${argc_filename}" -w "%{http_code}" \
+	model=$(echo ${voices_model[$argc_voice]})
+	http_status_code=$(curl -s -X POST "${tts_host}/v1/audio/speech" -o "${argc_filename}" -w "%{http_code}" \
 		-H "Authorization: Bearer ${LLM_API_KEY}" \
 		-H "Content-Type: application/json" \
-		-d "{\"model\": \"tts-1\",\"input\": \"${argc_text}\",\"voice\": \"${argc_voice}\",\"response_format\": \"wav\",\"speed\": ${argc_speed}}")
+		-d "{\"model\": \"${model}\",\"input\": \"${argc_text}\",\"voice\": \"${argc_voice}\",\"response_format\": \"wav\",\"speed\": ${argc_speed}}")

 	# Check the response code for successful HTTP request
 	if [[ "${http_status_code}" -ne 200 ]]; then