更新

2026-11-09 10:02:27 +08:00
parent 0c43d54c08
commit 8466acef9f
35 changed files with 3584 additions and 111 deletions
@@ -0,0 +1,104 @@
+<?php
+
+declare(strict_types=1);
+
+namespace app\adminapi\service;
+
+use app\common\model\tcm\CallRecord;
+use DomainException;
+use think\facade\Db;
+
+/** Persist a locally recovered transcript without reopening the realtime session. */
+final class LocalAudioTranscriptRecovery
+{
+    public static function save(array $params): array
+    {
+        foreach (['diagnosis_id', 'call_record_id', 'admin_id'] as $key) {
+            $value = $params[$key] ?? null;
+            if ((!is_int($value) && !is_string($value))
+                || filter_var($value, FILTER_VALIDATE_INT, ['options' => ['min_range' => 1]]) === false) {
+                throw new DomainException('录音补转写身份参数无效');
+            }
+        }
+        foreach (['recovery_id', 'expected_session_id', 'expected_status', 'audio_url', 'transcript_text'] as $key) {
+            if (!isset($params[$key]) || !is_string($params[$key])) {
+                throw new DomainException('录音补转写参数无效');
+            }
+        }
+        if (!preg_match('/\Aaudio-[a-f0-9]{32}\z/D', $params['recovery_id'])
+            || mb_strlen($params['expected_session_id']) > 128
+            || !in_array($params['expected_status'], ['', 'pending', 'running', 'partial', 'failed', 'completed'], true)
+            || trim($params['audio_url']) === '' || strlen($params['audio_url']) > 8192) {
+            throw new DomainException('录音补转写会话参数无效');
+        }
+        $text = trim($params['transcript_text']);
+        if ($text === '' || mb_strlen($params['transcript_text']) > 100000) {
+            throw new DomainException('录音补转写文字必须为1到100000个字符');
+        }
+
+        return Db::transaction(static function () use ($params, $text): array {
+            // The realtime upsert/finish endpoints lock the same row, so a late
+            // old-session write cannot run after this session switch succeeds.
+            $record = CallRecord::where('id', (int) $params['call_record_id'])
+                ->where('diagnosis_id', (int) $params['diagnosis_id'])
+                ->where('caller_id', (int) $params['admin_id'])
+                ->where('caller_type', 'doctor')->lock(true)->find();
+            if (!$record) { throw new DomainException('通话记录不存在或无权操作'); }
+            if ((int) $record['status'] !== 2) { throw new DomainException('通话结束后才能补转写'); }
+            $urls = json_decode((string) ($record['local_audio_urls'] ?? ''), true);
+            if ((int) ($record['local_audio_status'] ?? 0) !== 2 || !is_array($urls)
+                || !in_array($params['audio_url'], $urls, true)) {
+                throw new DomainException('录音尚未上传或不属于当前通话');
+            }
+            $result = ['call_record_id' => (int) $record['id'], 'transcription_status' => 'completed'];
+            $session = (string) ($record['transcription_session_id'] ?? '');
+            $status = (string) ($record['transcription_status'] ?? '');
+            if ($status === 'completed' && trim((string) ($record['transcript_text'] ?? '')) !== '') {
+                // Lost response retries succeed even with the original CAS values;
+                // a different completed session always wins without being changed.
+                return $session === $params['recovery_id']
+                    ? $result + ['saved' => true]
+                    : $result + ['saved' => false, 'already_complete' => true];
+            }
+            if ($session !== $params['expected_session_id'] || $status !== $params['expected_status']) {
+                throw new DomainException('通话转写状态已变化，请刷新后重试');
+            }
+
+            $now = time();
+            $segments = [];
+            $lines = [];
+            // Recognition supplied no diarization or alignment. Do not invent
+            // doctor/patient attribution or timestamps; preserve source order by id.
+            for ($offset = 0, $index = 0, $length = mb_strlen($text); $offset < $length; $offset += 4000, $index++) {
+                $part = mb_substr($text, $offset, 4000);
+                $segments[] = [
+                    'call_record_id' => (int) $record['id'],
+                    'transcription_session_id' => $params['recovery_id'],
+                    'segment_id' => $params['recovery_id'] . '-' . $index,
+                    'speaker_user_id' => '', 'speaker_role' => 'unknown', 'timestamp_ms' => 0,
+                    'text' => $part, 'create_time' => $now, 'update_time' => $now,
+                ];
+                $lines[] = '未知说话人：' . $part;
+            }
+            // Keep all old-session segments for audit; the current session is the
+            // only one consumed by transcript readers and clinical context builders.
+            Db::name('tcm_call_transcript_segment')->insertAll($segments);
+            $record->save([
+                'transcription_session_id' => $params['recovery_id'],
+                'transcription_status' => 'completed',
+                'transcription_segment_count' => count($segments),
+                'transcript_text' => implode("\n", $lines),
+                'transcription_started_at' => $now, 'transcription_finished_at' => $now,
+                'update_time' => $now,
+            ]);
+            return $result + ['saved' => true];
+        });
+    }
+
+    public static function completionStatus(string $requested, int $actual, int $expected, string $text): string
+    {
+        if ($requested !== 'completed') { return $requested; }
+        if ($actual === 0 || trim($text) === '') { return 'failed'; }
+        return $actual < $expected ? 'partial' : 'completed';
+    }
+}