Merge pull request #11 from SEPIA-Framework/dev

v1.0.0 release candidate

Author: fquirin

.gitignore

@@ -140,3 +140,4 @@ cython_debug/
 # PROJECT
 recordings/
 models/
+src/server_test.conf

CHANGELOG.md

@@ -0,0 +1,19 @@
+# Changelog for SEPIA Speech-To-Text Server
+
+## v1.0.0 - xx.xx.xxxx
+
+- Added support for Coqui-STT and fully implemented hot-swapping of engines (currently: Vosk + Coqui)
+- Added new config file properties for models: 'engine', 'task', 'scorer', 'name'
+- Added new config file options for 'asr_engine': 'coqui' and 'dynamic'
+- Reworked config file loader to load only ONE file with all settings (no overwrites)
+- Added new common 'welcome' message options 'task' and Coqui specific 'hotWords' and 'scorer'
+- Accept alias names in 'welcome' message (streamlined with features): 'words_ts', 'phrase_list', 'hot_words' and 'external_scorer'
+- Reworked engine interface to load best model depending on: name, full language code (e.g.: de-DE), partial language code (e.g.: de), task or defaults
+- Added WebSocket connection heartbeat and timeout to config file
+- Improved error handling
+- Improved Vosk test script and added Coqui test
+- Updated HTML test and demo page
+
+## v0.9.5 - 08.05.2021
+
+- Release of new STT-Server

README.md

@@ -27,26 +27,33 @@ If you are using custom models built for the 2018 version you can easily [conver
 
 ## Integrated ASR Engines
 
-- [Vosk](https://github.com/alphacep/vosk-api) - Status: Ready (Note: non-CUDA build). Includes tiny EN and DE models.
-- [Coqui](https://github.com/coqui-ai/STT) - Status: Planned.
+- [Vosk](https://github.com/alphacep/vosk-api) - Status: **Ready**. Includes tiny EN and DE models.
+- [Coqui](https://github.com/coqui-ai/STT) - Status: **Ready**. Includes basic English model w/o scorer.
 - [Scribosermo](https://gitlab.com/Jaco-Assistant/Scribosermo) - Status: [experimenting with setup](https://github.com/fquirin/scribosermo-stt-setup).
+- [Nvidia NeMo](https://github.com/NVIDIA/NeMo) - Status: experimenting with setup.
 - [TensorFlowASR](https://github.com/TensorSpeech/TensorFlowASR) - Status: Help wanted.
-- If you want to see additional engines please create a new [issue](https://github.com/SEPIA-Framework/sepia-stt-server/issues). Pull requests are welcome ;-)
+
+If you want to see additional engines please create a new [issue](https://github.com/SEPIA-Framework/sepia-stt-server/issues). Pull requests are welcome ;-)
 
 ## Quick-Start with Python
 
 The easiest way to get started is to use a Docker container for your platform. To install the server yourself please see the [code section README](src/README.md).
 
 ## Quick-Start with Docker
 
-Choose your platform and pull the image. The smallest English and German Vosk models are already included:
-- ARM 32Bit (Raspberry Pi 4 32Bit OS): `docker pull sepia/stt-server:vosk_armv7l`
-- ARM 64Bit (RPi 4 64Bit, Jetson Nano(?)): `docker pull sepia/stt-server:vosk_aarch64`
-- x86 64Bit Systems (Desktop PCs, Linux server etc.): `docker pull sepia/stt-server:vosk_amd64`
+Simply pull the latest image (or choose an older one from the archive). The smallest English and German Vosk models and an English Coqui model (w/o scorer) are included:
+```
+`docker pull sepia/stt-server:latest`
+```
+
+Supported platforms:
+- ARM 32Bit (Raspberry Pi 4 32Bit OS) 
+- ARM 64Bit (RPi 4 64Bit, Jetson Nano(?))
+- x86 64Bit Systems (Desktop PCs, Linux server etc.)
 
-After the download is complete simply start the container, for example via:  
+After the download is finished you can start the container like this:  
 ```
-sudo docker run --rm --name=sepia-stt -p 20741:20741 -it sepia/stt-server:[image-tag]
+sudo docker run --rm --name=sepia-stt -p 20741:20741 -it sepia/stt-server:latest
 ```
 
 ### Test via web interface
@@ -55,18 +62,19 @@ To test the server visit: `http://localhost:20741` if you are on the same machin
 
 ### Models
 
-Currently the server supports [Vosk ASR models](https://alphacephei.com/vosk/models) and custom models (see "adapt" section below).  
+Currently the server supports [Vosk ASR models](https://alphacephei.com/vosk/models), [Coqui-STT models](https://coqui.ai/models) and custom models (see "adapt" section below).  
 
 To **add new ASR models** create a shared volume for your container, place your model inside and update the server [config file](src/server.conf). The "adapt" section below has a more detailed example, but basically you can:
 - Add a volume to your container, e.g. use run flag: `-v [host-models-folder]:/home/admin/sepia-stt/models/my` (Note: use absolute path!)
 - Copy your model folder (e.g. 'vosk-model-small-es') and the server [config file](src/server.conf) to your new folder
-- Add model path and language code to the "[asr_models]" section in your config, e.g.: `path3=my/vosk-model-small-es` and `lang3=es-ES`
+- Add model path and language code to the "[asr_models]" section in your config, e.g.: `path3=my/vosk-model-small-es`, `lang3=es-ES`, `engine3=vosk` and optionally a "task" like `task3=smart-home`
 - Tell the server to use your new config via the flag: `--env SEPIA_STT_SETTINGS=/home/admin/sepia-stt/models/my/server.conf`
 
 Included inside the Docker containers are:
-- [vosk-model-small-en-us-0.15](https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-small-en-us-0.15.zip) - Lightweight wideband model for Android and RPi - Apache 2.0 license
-- [vosk-model-small-de-0.15](https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-small-de-0.15.zip) - Lightweight wideband model for Android and RPi - Apache 2.0 license
-- [vosk-model-spk-0.4](https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-spk-0.4.zip) - Model for speaker identification (all models) - Apache 2.0 license
+- [vosk-model-small-en-us-0.15](https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-small-en-us-0.15.zip) - Vosk lightweight wideband model for Android and RPi - Apache 2.0 license
+- [vosk-model-small-de-0.15](https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-small-de-0.15.zip) - Vosk lightweight wideband model for Android and RPi - Apache 2.0 license
+- [vosk-model-spk-0.4](https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-spk-0.4.zip) - Vosk model for speaker identification (all models) - Apache 2.0 license
+- [coqui-model-en-1.0.0](https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/coqui-model-en-1.0.0.zip) - Coqui basic English model w/o scorer for testing - Apache 2.0 license
 
 ## Server Settings
 
@@ -75,15 +83,15 @@ Most of the settings can be handled easily via the [server.conf settings file](s
 ENV variables:
 - `SEPIA_STT_SETTINGS`: Overwrites default path to settings file
 
-Commandline options:
-- Use `python -m launch -h` to see all commandline options
+Command line options:
+- Use `python -m launch -h` to see all command line options
 - Use `python -m launch -s [path-to-file]` to use custom settings
 
-NOTE: Commandline options always overrule the settings file but in most scenarios it makes sense to simply create a new settings file and use the `-s` flag.
+NOTE: Command line options always overrule the settings file but in most scenarios it makes sense to simply create a new settings file and use the `-s` flag.
 
 ## ASR Engine Settings
 
-As soon as the server is running you can check the current setup via the HTTP REST interface: `http://localhost:20741//settings` or the test page (see quick-start above).  
+As soon as the server is running you can check the current setup via the HTTP REST interface: `http://localhost:20741/settings` or the test page (see quick-start above).  
 
 Individual settings for the active engine can be changed on-the-fly during the WebSocket 'welcome' event. See the [API docs](API.md) file for more info or check out the 'Engine Settings' section of the test page.
 
@@ -129,7 +137,7 @@ sudo docker run --rm --name=sepia-stt -p 20741:20741 -it \
 	-v [host-models-folder]:/home/admin/sepia-stt/models/my \
 	-v [host-share-folder]:/home/admin/sepia-stt/share \
 	--env SEPIA_STT_SETTINGS=/home/admin/sepia-stt/share/my.conf \
-	sepia/stt-server:[image-tag] \
+	sepia/stt-server:latest \
 	/bin/bash
 ```
 

engines/dynamic/Dockerfile

@@ -0,0 +1,122 @@
+FROM debian:buster-slim
+
+# Default to UTF-8 file.encoding
+ENV LANG C.UTF-8
+
+# Run 1
+RUN echo 'Installing dependencies...' && \
+#
+#   Dependencies
+    apt-get update && \
+    apt-get install -y --no-install-recommends \
+    sudo git wget curl nano unzip zip procps \
+    build-essential \
+    python3-pip python3-dev python3-setuptools python3-wheel \
+    #amd64 only? moved to adapt package: libatlas-base-dev \
+    libffi-dev && \
+    pip3 install pip --upgrade && \
+#
+#   Fast-API
+    pip3 install cffi && \
+    pip3 install fastapi uvicorn[standard] aiofiles && \
+#
+#   Clean up
+    apt-get remove -y build-essential && \
+    apt-get install libatomic1 && \
+    apt-get clean && apt-get autoclean && apt-get autoremove -y && \
+#
+#   Create user
+    useradd --create-home --shell /bin/bash admin && \
+    adduser admin sudo && \
+    echo "admin ALL=(ALL) NOPASSWD: ALL" >> /etc/sudoers
+#
+#   ENV
+#SOME_ENV_VAR=/...my-stuff
+
+#   USER
+USER admin
+
+# Run 1
+RUN echo "Installing Vosk ..." && \
+	mkdir -p /home/admin/install && \
+	mkdir -p /home/admin/sepia-stt/models && \
+	mkdir -p /home/admin/kaldi-adapt-lm && \
+	cd /home/admin/install && \
+	if [ -n "$(uname -m | grep aarch64)" ]; then \
+		echo "Downloading Vosk 0.3.42 for aarch64"; \
+		#wget https://github.com/alphacep/vosk-api/releases/download/v0.3.42/vosk-0.3.42-py3-none-linux_aarch64.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/vosk-0.3.42-py3-none-linux_aarch64.whl; \
+		pip3 install vosk-0.3.42-py3-none-linux_aarch64.whl; \
+	elif [ -n "$(uname -m | grep armv7l)" ]; then \
+		echo "Downloading Vosk 0.3.42 for armv7l"; \
+		#wget https://github.com/alphacep/vosk-api/releases/download/v0.3.42/vosk-0.3.42-py3-none-linux_armv7l.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/vosk-0.3.42-py3-none-linux_armv7l.whl; \
+		pip3 install vosk-0.3.42-py3-none-linux_armv7l.whl; \
+	else \
+		echo "Downloading Vosk 0.3.42 for x86_64"; \
+		#wget https://github.com/alphacep/vosk-api/releases/download/v0.3.42/vosk-0.3.42-py3-none-linux_x86_64.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/vosk-0.3.42-py3-none-linux_x86_64.whl; \
+		pip3 install vosk-0.3.42-py3-none-linux_x86_64.whl; \
+	fi && \
+	#wget https://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip && \
+	#wget https://alphacephei.com/vosk/models/vosk-model-small-de-0.15.zip && \
+	#wget https://alphacephei.com/vosk/models/vosk-model-spk-0.4.zip && \
+	wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-small-en-us-0.15.zip && \
+	wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-small-de-0.15.zip && \
+	wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-model-spk-0.4.zip && \
+	unzip vosk-model-small-en-us-0.15.zip && \
+	mv vosk-model-small-en-us-0.15 /home/admin/sepia-stt/models/vosk-model-small-en-us && \
+	unzip vosk-model-small-de-0.15.zip && \
+	mv vosk-model-small-de-0.15 /home/admin/sepia-stt/models/vosk-model-small-de && \
+	unzip vosk-model-spk-0.4.zip && \
+	mv vosk-model-spk-0.4 /home/admin/sepia-stt/models/vosk-model-spk && \
+#
+	echo "Installing Coqui-STT ..." && \
+	mkdir -p /home/admin/sepia-stt/models/coqui-model-en && \
+	if [ -n "$(uname -m | grep aarch64)" ]; then \
+		echo "Downloading Coqui-STT 1.3.0 for aarch64"; \
+		#wget https://github.com/coqui-ai/STT/releases/download/v1.3.0/stt-1.3.0-cp37-cp37m-linux_aarch64.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/stt-1.3.0-cp37-cp37m-linux_aarch64.whl; \
+		pip3 install stt-1.3.0-cp37-cp37m-linux_aarch64.whl; \
+	elif [ -n "$(uname -m | grep armv7l)" ]; then \
+		echo "Downloading Coqui-STT 1.3.0 for armv7l"; \
+		#wget https://github.com/coqui-ai/STT/releases/download/v1.3.0/stt-1.3.0-cp37-cp37m-linux_armv7l.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/stt-1.3.0-cp37-cp37m-linux_armv7l.whl; \
+		pip3 install stt-1.3.0-cp37-cp37m-linux_armv7l.whl; \
+	else \
+		echo "Downloading Coqui-STT 1.3.0 for x86_64"; \
+		#wget https://github.com/coqui-ai/STT/releases/download/v1.3.0/stt-1.3.0-cp37-cp37m-manylinux_2_24_x86_64.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/stt-1.3.0-cp37-cp37m-manylinux_2_24_x86_64.whl; \
+		pip3 install stt-1.3.0-cp37-cp37m-manylinux_2_24_x86_64.whl; \
+	fi && \
+	#cd /home/admin/sepia-stt/models/coqui-model-en && \
+	#wget https://coqui.gateway.scarf.sh/english/coqui/v1.0.0-large-vocab/model.tflite && \
+	#cd /home/admin/install && \
+	wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/coqui-model-en-1.0.0.zip && \
+	unzip coqui-model-en-1.0.0.zip && \
+	mv coqui-model-en-1.0.0 /home/admin/sepia-stt/models/coqui-model-en && \
+#
+	echo "Installing SEPIA STT ..." && \
+	SEPIA_STT_BRANCH=master && \
+	git clone --single-branch --depth 1 -b $SEPIA_STT_BRANCH https://github.com/SEPIA-Framework/sepia-stt-server.git && \
+	mv sepia-stt-server/src /home/admin/sepia-stt/server && \
+#
+	echo "Installing Adapt-LM scripts ..." && \
+	cd /home/admin && \
+	ADAPT_LM_BRANCH=master && \
+	git clone --single-branch --depth 1 -b $ADAPT_LM_BRANCH https://github.com/fquirin/kaldi-adapt-lm.git && \
+	cd kaldi-adapt-lm && \
+	bash 1-download-requirements.sh && \
+	rm *.tar.gz && \
+#
+#	Clean up install folder
+	cd /home/admin && \
+	sudo rm -rf /home/admin/install && \
+#
+# TODO: install proxy with self-signed certs?
+#
+	echo "#!/bin/bash" > on-docker.sh && echo "cd sepia-stt/server && python3 -m launch" >> on-docker.sh
+	
+# Start
+WORKDIR /home/admin
+CMD bash on-docker.sh

engines/dynamic/build_container.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+version=latest
+if [ -n "$1" ]; then
+	version=$1
+fi
+if [ -n "$(uname -m | grep aarch64)" ]; then
+	echo "Building 'dynamic' Docker container for $version_aarch64"
+	sudo docker build --no-cache -t sepia/stt-server:"dynamic_$version_aarch64" .
+elif [ -n "$(uname -m | grep armv7l)" ]; then
+	echo "Building 'dynamic' Docker container for $version_armv7l"
+	sudo docker build --no-cache -t sepia/stt-server:"dynamic_$version_armv7l" .
+else
+	# NOTE: x86 32bit build not supported atm
+	echo "Building 'dynamic' Docker container for $version_amd64"
+	sudo docker build --no-cache -t sepia/stt-server:"dynamic_$version_amd64" .
+fi

engines/dynamic/create_docker_manifest.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+version=latest
+if [ -n "$1" ]; then
+	version=$1
+fi
+sudo docker manifest create sepia/stt-server:latest \
+--amend "sepia/stt-server:dynamic_$version_aarch64" \
+--amend "sepia/stt-server:dynamic_$version_armv7l" \
+--amend "sepia/stt-server:dynamic_$version_amd64"
+
+sudo docker manifest push sepia/stt-server:latest

engines/dynamic/test_basic_server.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+IMAGE_TAG=latest
+sudo docker run --rm --name=sepia-stt-vosk -p 20741:20741 -it sepia/stt-server:$IMAGE_TAG /bin/bash
+#-v share:/home/admin/share

engines/dynamic/test_terminal_with_vol.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+IMAGE_TAG=latest
+HOST_MODELS="$(realpath ~)/stt/models"
+HOST_SHARE="$(realpath ~)/stt/share"
+mkdir -p "$HOST_MODELS"
+mkdir -p "$HOST_SHARE"
+sudo docker run --rm --name=sepia-stt -p 20741:20741 -it \
+	-v "$HOST_MODELS":/home/admin/sepia-stt/models/my \
+	-v "$HOST_SHARE":/home/admin/sepia-stt/share \
+	--env SEPIA_STT_SETTINGS=/home/admin/sepia-stt/share/my.conf \
+	sepia/stt-server:$IMAGE_TAG \
+	/bin/bash

engines/vosk/Dockerfile

@@ -15,7 +15,7 @@ RUN echo 'Installing dependencies...' && \
     #amd64 only? moved to adapt package: libatlas-base-dev \
     libffi-dev && \
 #
-#   Vosk and Fast-API
+#   Fast-API
     pip3 install cffi && \
     pip3 install fastapi uvicorn[standard] aiofiles && \
 #
@@ -42,20 +42,20 @@ RUN echo "Installing Vosk ..." && \
 	mkdir -p /home/admin/kaldi-adapt-lm && \
 	cd /home/admin/install && \
 	if [ -n "$(uname -m | grep aarch64)" ]; then \
-		echo "Downloading Vosk 0.3.30 for aarch64"; \
-		#wget https://github.com/alphacep/vosk-api/releases/download/0.3.30/vosk-0.3.30-py3-none-linux_aarch64.whl; \
-		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-0.3.30-py3-none-linux_aarch64.whl; \
-		pip3 install vosk-0.3.30-py3-none-linux_aarch64.whl; \
+		echo "Downloading Vosk 0.3.42 for aarch64"; \
+		#wget https://github.com/alphacep/vosk-api/releases/download/v0.3.42/vosk-0.3.42-py3-none-linux_aarch64.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/vosk-0.3.42-py3-none-linux_aarch64.whl; \
+		pip3 install vosk-0.3.42-py3-none-linux_aarch64.whl; \
 	elif [ -n "$(uname -m | grep armv7l)" ]; then \
-		echo "Downloading Vosk 0.3.30 for armv7l"; \
-		#wget https://github.com/alphacep/vosk-api/releases/download/0.3.30/vosk-0.3.30-py3-none-linux_armv7l.whl; \
-		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-0.3.30-py3-none-linux_armv7l.whl; \
-		pip3 install vosk-0.3.30-py3-none-linux_armv7l.whl; \
+		echo "Downloading Vosk 0.3.42 for armv7l"; \
+		#wget https://github.com/alphacep/vosk-api/releases/download/v0.3.42/vosk-0.3.42-py3-none-linux_armv7l.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/vosk-0.3.42-py3-none-linux_armv7l.whl; \
+		pip3 install vosk-0.3.42-py3-none-linux_armv7l.whl; \
 	else \
-		echo "Downloading Vosk 0.3.30 for x86_64"; \
-		#wget https://github.com/alphacep/vosk-api/releases/download/0.3.30/vosk-0.3.30-py3-none-linux_x86_64.whl; \
-		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v0.9.5/vosk-0.3.30-py3-none-linux_x86_64.whl; \
-		pip3 install vosk-0.3.30-py3-none-linux_x86_64.whl; \
+		echo "Downloading Vosk 0.3.42 for x86_64"; \
+		#wget https://github.com/alphacep/vosk-api/releases/download/v0.3.42/vosk-0.3.42-py3-none-linux_x86_64.whl; \
+		wget https://github.com/SEPIA-Framework/sepia-stt-server/releases/download/v1.0.0/vosk-0.3.42-py3-none-linux_x86_64.whl; \
+		pip3 install vosk-0.3.42-py3-none-linux_x86_64.whl; \
 	fi && \
 	#wget https://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip && \
 	#wget https://alphacephei.com/vosk/models/vosk-model-small-de-0.15.zip && \

engines/vosk/build_container.sh

@@ -1,13 +1,12 @@
 #!/bin/bash
-# TODO: make version number variable - add '--no-cache' ?
 if [ -n "$(uname -m | grep aarch64)" ]; then
 	echo "Building Vosk Docker container for aarch64"
-	sudo docker build -t sepia/stt-server:vosk_aarch64 .
+	sudo docker build --no-cache -t sepia/stt-server:vosk_aarch64 .
 elif [ -n "$(uname -m | grep armv7l)" ]; then
 	echo "Building Vosk Docker container for armv7l"
-	sudo docker build -t sepia/stt-server:vosk_armv7l .
+	sudo docker build --no-cache -t sepia/stt-server:vosk_armv7l .
 else
 	# NOTE: x86 32bit build not supported atm
 	echo "Building Vosk Docker container for amd64"
-	sudo docker build -t sepia/stt-server:vosk_amd64 .
+	sudo docker build --no-cache -t sepia/stt-server:vosk_amd64 .
 fi