SubtitleEdit/subtitleedit · error · InvalidOperationException

IndexTTS (CrispASR) — the crispasr server crashed during syn

Error message

IndexTTS (CrispASR) — the crispasr server crashed during synthesis.

What it means

InvalidOperationException from the HttpRequestException catch in IndexTtsCrispAsr.Speak when the POST to `/v1/audio/speech` throws AND the server process has already exited (`died == true`). The server crashed mid-synthesis; StopServerInternal is called, the inner HttpRequestException is chained, and the server log + launch command are appended. This means a fresh server will be started on the next call.

Source

Thrown at src/ui/Features/Video/TextToSpeech/Engines/IndexTtsCrispAsr.cs:424

            response = await HttpClient.PostAsync($"{ServerBaseUrl}/v1/audio/speech", content, cancellationToken);
        }
        catch (HttpRequestException ex)
        {
            var serverLog = SnapshotServerLog();
            var launchCommand = _serverLaunchCommand;
            var died = _serverProcess?.HasExited == true;
            if (died)
            {
                StopServerInternal();
            }

            var failMsg = $"IndexTTS (CrispASR) request failed — Voice: {indexVoice}, Text: {text}, "
                + $"RequestJson: {body}, ServerExited: {died}, ServerLog: {serverLog}"
                + LaunchCmdSuffix(launchCommand);
            Se.LogError(ex, failMsg);
            Se.WriteToolsLog(failMsg);

            throw new InvalidOperationException(
                (died
                    ? "IndexTTS (CrispASR) — the crispasr server crashed during synthesis."
                    : "IndexTTS (CrispASR) request failed — the connection to the crispasr server was dropped.")
                + (string.IsNullOrEmpty(serverLog) ? string.Empty : $"{Environment.NewLine}Server log:{Environment.NewLine}{serverLog}")
                + LaunchCmdSuffix(launchCommand),
                ex);
        }

        using (response)
        {
            if (!response.IsSuccessStatusCode)
            {
                var errorBody = await SafeReadErrorAsync(response, cancellationToken);
                var serverLog = SnapshotServerLog();
                var launchCommand = _serverLaunchCommand;
                var errMsg = $"IndexTTS (CrispASR) server error {(int)response.StatusCode} {response.StatusCode} — "
                    + $"Voice: {indexVoice}, Text: {text}, RequestJson: {body}, "
                    + $"ResponseBody: {errorBody}, ServerLog: {serverLog}"

View on GitHub (pinned to 17a9f07487)

Solutions

Inspect the appended `Server log:` tail — the backend's crash reason is logged right before exit.
Retry once; EnsureServerRunningAsync will start a fresh server. If it dies again on the same input, the reference WAV or model is the cause.
Re-encode the reference WAV to 24 kHz mono PCM.
Check VRAM/free RAM at the moment of failure; close other GPU workloads.
If the crash is input-specific (long text), chunk the text and retry per chunk.

Defensive patterns

Strategy: retry

Validate before calling

// Before batch synth, probe the server is alive and the reference WAV is valid
if (engine.IsServerDown() || !File.Exists(indexVoice.FilePath)) { WarnUser(...); return; }

Try / catch

try { await indexEngine.Speak(...); }
catch (InvalidOperationException ex) when (ex.Message.Contains("server crashed during synthesis"))
{
    // Server is auto-stopped; EnsureServerRunningAsync will start a fresh one. Retry once.
    await indexEngine.Speak(...);
}

Prevention

Pre-validate the reference WAV format (24 kHz mono) to avoid mid-synth segfaults.
Monitor VRAM and skip synth while another GPU job is running.
Keep batch jobs single-threaded per server to avoid concurrent-clone crashes.
Log the inner HttpRequestException to correlate with backend stack traces.

When it happens

Trigger: The reference WAV is malformed so the backend segfaults during cloning; an OOM/GPU fault kills the server mid-inference; the model GGUF is corrupt and the server aborts on the first synth; a CUDA driver crash during generation.

Common situations: Reference WAV is the wrong sample rate/channel layout; GPU VRAM exhausted by another process mid-render; corrupt or truncated model download; CrispASR backend bug on specific input text (e.g. very long or non-UTF8).

Related errors

AI-assisted analysis of SubtitleEdit/subtitleedit@17a9f07487 (2026-08-13). Data as JSON: /api/errors/d74d0378e48b166a. Report an issue: GitHub.