Merge pull request #2244 from CapSoftware/codex/fix-transcription-stream-probe

fix: recover valid audio probes and classify no-speech transcription

Author: richiemcilroy

apps/media-server/Dockerfile

@@ -15,6 +15,7 @@ RUN bun test src/__tests__/lib/audio-quality-policy.test.ts src/__tests__/lib/au
 	&& bun test src/__tests__/lib/audio-quality-formats.integration.test.ts src/__tests__/lib/audio-quality-benchmark.test.ts \
 	&& bun test src/__tests__/lib/drive-resumable-upload.test.ts src/__tests__/lib/storage-upload.test.ts src/__tests__/lib/container-memory.test.ts \
 	&& bun test src/__tests__/lib/media-size.test.ts src/__tests__/lib/media-probe.integration.test.ts \
+	&& bun test src/__tests__/lib/media-audio.integration.test.ts \
 	&& bun test src/__tests__/lib/recording-verification.integration.test.ts src/__tests__/lib/job-manager.test.ts \
 	&& bun test src/__tests__/lib/media-transfer.test.ts \
 	&& bun test src/__tests__/routes/recording-verification.test.ts \

apps/media-server/src/__tests__/lib/media-audio.integration.test.ts

@@ -104,6 +104,48 @@ describe("mediaAudio integration tests", () => {
 			expect(hasAudio).toBe(false);
 		});
 
+		test.each([
+			[90, true],
+			[-90, true],
+			[180, true],
+			[90, false],
+		] as const)(
+			"detects audio with display rotation %i and audio=%s",
+			async (rotation, hasAudio) => {
+				const dirPath = await mkdtemp(join(tmpdir(), "cap-audio-rotation-"));
+				const outputPath = join(dirPath, "rotated.mp4");
+				try {
+					const proc = Bun.spawn({
+						cmd: [
+							"ffmpeg",
+							"-v",
+							"error",
+							"-display_rotation",
+							String(rotation),
+							"-i",
+							hasAudio ? TEST_VIDEO_WITH_AUDIO : TEST_VIDEO_NO_AUDIO,
+							"-c",
+							"copy",
+							outputPath,
+						],
+						stdout: "ignore",
+						stderr: "pipe",
+					});
+					const [stderr, exitCode] = await Promise.all([
+						new Response(proc.stderr).text(),
+						proc.exited,
+					]);
+					expect(stderr).toBe("");
+					expect(exitCode).toBe(0);
+					expect(await checkHasAudioTrack(`file://${outputPath}`)).toBe(
+						hasAudio,
+					);
+				} finally {
+					await rm(dirPath, { recursive: true, force: true });
+				}
+			},
+		);
+
 		test("inherits signed queries for relative HLS segments", async () => {
 			const dirPath = await createHlsFixture();
 			const requests: string[] = [];
@@ -134,6 +176,21 @@ describe("mediaAudio integration tests", () => {
 			}
 		});
 
+		test("still rejects audio-only inputs without leaking an operation", async () => {
+			const beforeCount = getActiveProcessCount();
+			const dirPath = await mkdtemp(join(tmpdir(), "cap-audio-only-"));
+			const outputPath = join(dirPath, "audio.mp3");
+			try {
+				await Bun.write(outputPath, await extractAudio(TEST_VIDEO_WITH_AUDIO));
+				await expect(
+					checkHasAudioTrack(`file://${outputPath}`),
+				).rejects.toThrow("No video stream found");
+				await waitForAudioOperations(beforeCount);
+			} finally {
+				await rm(dirPath, { recursive: true, force: true });
+			}
+		});
+
 		test("contains upstream failures without leaking the active operation", async () => {
 			const beforeCount = getActiveProcessCount();
 			const server = Bun.serve({

apps/media-server/src/lib/media-audio.ts

@@ -1,6 +1,7 @@
 import { mkdtemp, rm } from "node:fs/promises";
 import { join } from "node:path";
 import { type Subprocess, spawn } from "bun";
+import { z } from "zod";
 import { withTimeout } from "./media-common";
 import {
 	canAcceptNewAudioOperation,
@@ -31,6 +32,9 @@ const MAX_AUDIO_SIZE_BYTES = 100 * 1024 * 1024;
 const MAX_STDERR_BYTES = 64 * 1024;
 const AUDIO_PROBE_MAX_ATTEMPTS = 3;
 const AUDIO_PROBE_RETRY_BASE_MS = 250;
+const AUDIO_PROBE_SCHEMA = z.object({
+	streams: z.array(z.object({ codec_type: z.string().optional() })),
+});
 
 const DEFAULT_OPTIONS: Required<AudioExtractionOptions> = {
 	format: "mp3",
@@ -144,7 +148,7 @@ function getAudioProbeArgs(inputPath: string): string[] {
 			"0",
 		);
 	}
-	args.push("-show_entries", "stream=codec_type", "-of", "csv=p=0", inputPath);
+	args.push("-show_entries", "stream=codec_type", "-of", "json", inputPath);
 	return args;
 }
 
@@ -224,10 +228,8 @@ async function probeAudioTracks(
 			const safeStderrText = redactProcessOutput(stderrText, sourceUrl);
 
 			if (exitCode === 0) {
-				const trackTypes = stdoutText
-					.split(/\r?\n/)
-					.map((value) => value.trim())
-					.filter(Boolean);
+				const { streams } = AUDIO_PROBE_SCHEMA.parse(JSON.parse(stdoutText));
+				const trackTypes = streams.map((stream) => stream.codec_type);
 				if (!trackTypes.includes("video")) {
 					throw new Error("No video stream found");
 				}

apps/web/__tests__/integration/transcribe-workflow.test.ts

@@ -98,9 +98,12 @@ vi.mock("drizzle-orm", () => ({
 
 vi.mock("server-only", () => ({}));
 
-vi.mock("workflow", () => ({
-	FatalError: class FatalError extends Error {},
-}));
+vi.mock("workflow", async () => {
+	const { runInNewContext } = await import("node:vm");
+	return {
+		FatalError: runInNewContext("(class FatalError extends Error {})"),
+	};
+});
 
 vi.mock("workflow/api", () => ({
 	start: vi.fn(),
@@ -245,7 +248,9 @@ describe("transcribeVideoWorkflow", () => {
 		expect(mocks.updates.at(-1)).toEqual({ transcriptionStatus: "COMPLETE" });
 	});
 
-	it("marks audio without speech as skipped without retrying transcription", async () => {
+	it("handles no-speech errors across workflow realms without retrying transcription", async () => {
+		const { FatalError } = await import("workflow");
+		expect(new FatalError("no spoken audio")).not.toBeInstanceOf(Error);
 		mocks.transcribe.mockResolvedValueOnce({
 			id: "silent-transcript",
 			status: "error",
@@ -268,6 +273,7 @@ describe("transcribeVideoWorkflow", () => {
 		expect(mocks.updates).toContainEqual({ transcriptionStatus: "NO_AUDIO" });
 		expect(mocks.updates).not.toContainEqual({ transcriptionStatus: "ERROR" });
 		expect(mocks.startAiGeneration).not.toHaveBeenCalled();
+		expect(mocks.deleteObject).toHaveBeenCalledTimes(1);
 	});
 
 	it("preserves transcription failures unrelated to missing speech", async () => {

apps/web/workflows/transcribe.ts

@@ -162,7 +162,10 @@ export async function transcribeVideoWorkflow(
 		await saveTranscription(videoId, userId, videoData.video, transcription);
 	} catch (error) {
 		if (
-			error instanceof Error &&
+			typeof error === "object" &&
+			error !== null &&
+			"message" in error &&
+			typeof error.message === "string" &&
 			error.message.toLowerCase().includes("no spoken audio")
 		) {
 			await markNoAudio(videoId);