131 lines
4.1 KiB
Bash
Executable File
131 lines
4.1 KiB
Bash
Executable File
#!/usr/bin/env bash
|
||
#
|
||
# 验证 MOSS-Transcribe-Diarize 转写服务：
|
||
# 1. 检查 GPU 与服务运行状态
|
||
# 2. 用测试音频发起 verbose_json 转写请求，校验响应契约和说话人标签
|
||
# 3. （可选）检查 oral-trainer-api 的集成状态
|
||
#
|
||
# 用法：
|
||
# ./verify.sh
|
||
# ./verify.sh --file /path/to/speech.wav
|
||
# ./verify.sh --url http://127.0.0.1:8001 --model OpenMOSS-Team/MOSS-Transcribe-Diarize
|
||
# ./verify.sh --api-url http://127.0.0.1:8000
|
||
|
||
set -euo pipefail
|
||
|
||
MOSS_URL="${MOSS_URL:-http://127.0.0.1:8001}"
|
||
MOSS_MODEL="${MOSS_MODEL:-OpenMOSS-Team/MOSS-Transcribe-Diarize}"
|
||
TEST_FILE=""
|
||
API_URL=""
|
||
|
||
usage() {
|
||
sed -n '2,9p' "$0" | sed 's/^# \{0,1\}//'
|
||
}
|
||
|
||
while [[ $# -gt 0 ]]; do
|
||
case "$1" in
|
||
--url) MOSS_URL="$2"; shift 2 ;;
|
||
--model) MOSS_MODEL="$2"; shift 2 ;;
|
||
--file) TEST_FILE="$2"; shift 2 ;;
|
||
--api-url) API_URL="$2"; shift 2 ;;
|
||
-h|--help) usage; exit 0 ;;
|
||
*) echo "未知参数: $1"; usage; exit 1 ;;
|
||
esac
|
||
done
|
||
|
||
echo "==> MOSS 地址: $MOSS_URL"
|
||
echo "==> 模型: $MOSS_MODEL"
|
||
|
||
echo "==> 检查 GPU"
|
||
if command -v nvidia-smi >/dev/null 2>&1; then
|
||
nvidia-smi --query-gpu=name,memory.total,memory.used,driver_version --format=csv,noheader | head -5
|
||
else
|
||
echo "警告: 未找到 nvidia-smi，无法确认 GPU 状态。"
|
||
fi
|
||
|
||
echo "==> 检查服务健康"
|
||
if curl -fsS --max-time 10 "$MOSS_URL/health" > /dev/null 2>&1; then
|
||
echo "OK: /health 返回 200"
|
||
elif curl -fsS --max-time 10 "$MOSS_URL/healthz" > /dev/null 2>&1; then
|
||
echo "OK: /healthz 返回 200"
|
||
else
|
||
echo "警告: /health 和 /healthz 均不可达，服务可能尚未就绪。"
|
||
fi
|
||
|
||
TMP_DIR="$(mktemp -d)"
|
||
trap 'rm -rf "$TMP_DIR"' EXIT
|
||
|
||
if [[ -n "$TEST_FILE" ]]; then
|
||
AUDIO_FILE="$TEST_FILE"
|
||
else
|
||
echo "==> 生成测试音频（3 秒 440Hz 正弦波）"
|
||
if ! command -v ffmpeg >/dev/null 2>&1; then
|
||
echo "错误: 未找到 ffmpeg，请安装或用 --file 指定真实语音文件。" >&2
|
||
exit 1
|
||
fi
|
||
AUDIO_FILE="$TMP_DIR/tone.wav"
|
||
ffmpeg -hide_banner -loglevel error -f lavfi \
|
||
-i "sine=frequency=440:duration=3" -ar 16000 -ac 1 -y "$AUDIO_FILE"
|
||
fi
|
||
|
||
if [[ ! -f "$AUDIO_FILE" ]]; then
|
||
echo "错误: 音频文件不存在: $AUDIO_FILE" >&2
|
||
exit 1
|
||
fi
|
||
|
||
echo "==> 发起转写请求（verbose_json）"
|
||
START_TS="$(date +%s)"
|
||
RESPONSE="$(curl -sS --max-time 600 \
|
||
-X POST "$MOSS_URL/v1/audio/transcriptions" \
|
||
-F "model=$MOSS_MODEL" \
|
||
-F "file=@$AUDIO_FILE" \
|
||
-F "response_format=verbose_json" \
|
||
-F "temperature=0" \
|
||
-F "max_new_tokens=65536")"
|
||
ELAPSED="$(( $(date +%s) - START_TS ))"
|
||
|
||
echo "$RESPONSE" | python3 -c '
|
||
import json, sys
|
||
|
||
payload = json.load(sys.stdin)
|
||
text = payload.get("text")
|
||
segments = payload.get("segments")
|
||
|
||
assert isinstance(text, str), "响应缺少 text 字段"
|
||
assert isinstance(segments, list), "响应缺少 segments 字段（需要 verbose_json）"
|
||
|
||
speakers = set()
|
||
for index, segment in enumerate(segments):
|
||
start = segment.get("start")
|
||
end = segment.get("end")
|
||
speaker = segment.get("speaker")
|
||
if not (isinstance(start, (int, float)) and isinstance(end, (int, float)) and end > start >= 0):
|
||
raise AssertionError(f"segments[{index}] 缺少合法 start/end")
|
||
if speaker:
|
||
speakers.add(speaker)
|
||
|
||
print(f"OK: text={text!r}")
|
||
print(f"OK: segments={len(segments)} 条")
|
||
if speakers:
|
||
print(f"OK: 说话人={sorted(speakers)}")
|
||
else:
|
||
print("注意: 未见 speaker 字段；单说话人或模型未输出标签（正常弦音无多人对话）。")
|
||
'
|
||
echo "==> 转写耗时: ${ELAPSED}s"
|
||
|
||
if [[ -n "$API_URL" ]]; then
|
||
echo "==> 检查 API 集成 ($API_URL/healthz)"
|
||
curl -sS --max-time 15 "$API_URL/healthz" | python3 -c '
|
||
import json, sys
|
||
|
||
payload = json.load(sys.stdin)
|
||
configured = payload.get("moss_configured")
|
||
if configured is not True:
|
||
raise SystemExit(f"错误: moss_configured={configured!r}，请检查 .env 中的 MOSS_TRANSCRIBE_URL")
|
||
print("OK: moss_configured=true")
|
||
'
|
||
fi
|
||
|
||
echo "全部检查通过。建议再用真实多人对话音频复核说话人分离效果："
|
||
echo " $0 --file /path/to/multi-speaker.wav"
|