feat: add tts with go-piper (#649)

Signed-off-by: mudler <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto 2023-06-22 17:53:10 +02:00 committed by GitHub
parent cc31c58235
commit a7bb029d23
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
14 changed files with 237 additions and 29 deletions

View file

@ -5,8 +5,13 @@ FROM golang:$GO_VERSION as requirements
ARG BUILD_TYPE
ARG CUDA_MAJOR_VERSION=11
ARG CUDA_MINOR_VERSION=7
ARG SPDLOG_VERSION="1.11.0"
ARG PIPER_PHONEMIZE_VERSION='1.0.0'
ARG TARGETARCH
ARG TARGETVARIANT
ENV BUILD_TYPE=${BUILD_TYPE}
ARG GO_TAGS="stablediffusion tts"
RUN apt-get update && \
apt-get install -y ca-certificates cmake curl patch
@ -23,6 +28,8 @@ RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \
; fi
ENV PATH /usr/local/cuda/bin:${PATH}
WORKDIR /build
# OpenBLAS requirements
RUN apt-get install -y libopenblas-dev
@ -30,19 +37,37 @@ RUN apt-get install -y libopenblas-dev
RUN apt-get install -y libopencv-dev && \
ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
# piper requirements
# Use pre-compiled Piper phonemization library (includes onnxruntime)
#RUN if echo "${GO_TAGS}" | grep -q "tts"; then \
RUN curl -L "https://github.com/gabime/spdlog/archive/refs/tags/v${SPDLOG_VERSION}.tar.gz" | \
tar -xzvf - && \
mkdir -p "spdlog-${SPDLOG_VERSION}/build" && \
cd "spdlog-${SPDLOG_VERSION}/build" && \
cmake .. && \
make -j8 && \
cmake --install . --prefix /usr && mkdir -p "lib/Linux-$(uname -m)" && \
cd /build && \
mkdir -p "lib/Linux-$(uname -m)/piper_phonemize" && \
curl -L "https://github.com/rhasspy/piper-phonemize/releases/download/v${PIPER_PHONEMIZE_VERSION}/libpiper_phonemize-${TARGETARCH}${TARGETVARIANT}.tar.gz" | \
tar -C "lib/Linux-$(uname -m)/piper_phonemize" -xzvf - && ls -liah /build/lib/Linux-$(uname -m)/piper_phonemize/ && \
cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/lib/. /lib64/ && \
cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/lib/. /usr/lib/ && \
cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/include/. /usr/include/
# \
# ; fi
FROM requirements as builder
ARG GO_TAGS=stablediffusion
ARG GO_TAGS="stablediffusion tts"
ENV GO_TAGS=${GO_TAGS}
ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
ENV NVIDIA_VISIBLE_DEVICES=all
WORKDIR /build
COPY . .
RUN make build
RUN ESPEAK_DATA=/build/lib/Linux-$(uname -m)/piper_phonemize/lib/espeak-ng-data make build
FROM requirements