run.sh

#!/usr/bin/env bash

# Copyright 2024 Jinchuan Tian, Carnegie Mellon University, jinchuat@andrew.cmu.edu
#  Apache 2.0  (http://www.apache.org/licenses/LICENSE-2.0)

# Build environment for SpeechLM infernece

log() {
    local fname=${BASH_SOURCE[1]##*/}
    echo -e "$(date '+%Y-%m-%dT%H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
}

stage=1
stop_stage=100

. utils/parse_options.sh

if [ ! -f activate_python.sh ]; then
    log "Please run the env.sh to install conda environment" && exit 1;
else
    . ./activate_python.sh
fi
export PYTHONPATH=${PYTHONPATH}:./UniAudio/codec

if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
    log "start SpeechLM inference for TTS task"
    python3 -m espnet2.bin.speechlm_inference \
        --ngpu 1 \
        --nbest 1 \
        --model_file exp/speechlm_asr_tts_1.7B/31_40epoch.pth \
        --train_config exp/speechlm_asr_tts_1.7B/config.yaml \
        --config conf/decode_tts.yaml \
        --data_path_and_name_and_type dump/raw_codec_ssl_tts_librispeech/test_clean/data.json,_,dataset_json \
        --output_dir exp/speechlm_asr_tts_1.7B/tts_inference
fi