add whisper

2026-08-16 15:52:40 +08:00
parent 6e4d93cea6
commit 53cbcde543
6 changed files with 323 additions and 9 deletions
--- a/whisper/verify.sh
+++ b/whisper/verify.sh
@@ -0,0 +1,118 @@
+#!/usr/bin/env bash
+#
+# 验证 Whisper 转写服务：
+#   1. 检查 GPU 与容器运行状态
+#   2. 用测试音频发起 verbose_json 转写请求，校验响应契约
+#   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://127.0.0.1:9000 --model Systran/faster-whisper-large-v3
+#   ./verify.sh --api-url http://127.0.0.1:8000
+
+set -euo pipefail
+
+WHISPER_URL="${WHISPER_URL:-http://127.0.0.1:9000}"
+WHISPER_MODEL="${WHISPER_MODEL:-Systran/faster-whisper-large-v3}"
+TEST_FILE=""
+API_URL=""
+
+usage() {
+    sed -n '2,9p' "$0" | sed 's/^# \{0,1\}//'
+}
+
+while [[ $# -gt 0 ]]; do
+    case "$1" in
+        --url) WHISPER_URL="$2"; shift 2 ;;
+        --model) WHISPER_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 "==> Whisper 地址: $WHISPER_URL"
+echo "==> 模型: $WHISPER_MODEL"
+
+echo "==> 检查 GPU"
+if command -v nvidia-smi >/dev/null 2>&1; then
+    nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv,noheader | head -5
+else
+    echo "警告: 未找到 nvidia-smi，请确认 NVIDIA 驱动与容器运行时已安装（容器需要 --gpus 支持）。"
+fi
+
+echo "==> 检查容器状态"
+if command -v docker >/dev/null 2>&1 && [[ -f docker-compose.yml ]]; then
+    docker compose ps --status running | sed -n '1,3p'
+else
+    echo "警告: 未在当前目录发现 docker-compose.yml，跳过容器状态检查。"
+fi
+
+TMP_DIR="$(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="$(curl -sS --max-time 300 \
+    -X POST "$WHISPER_URL/v1/audio/transcriptions" \
+    -F "model=$WHISPER_MODEL" \
+    -F "file=@$AUDIO_FILE" \
+    -F "response_format=verbose_json" \
+    -F "temperature=0")"
+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）"
+
+for index, segment in enumerate(segments):
+    start = segment.get("start")
+    end = segment.get("end")
+    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")
+
+print(f"OK: text={text!r}")
+print(f"OK: segments={len(segments)} 条")
+'
+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。建议再用真实语音文件复核转写质量："
+echo "  $0 --file /path/to/speech.wav"