Skip to content

[bug] speech synthesize --timeout 180 does not terminate stalled audio streaming #309

Open

Description

@luodalu0003

Environment

  • CLI / bailian-gen / bailian-protocol: 2.8.4 (updated from 2.8.3 using the official updater before the run)
  • Platform: macOS 27.2, arm64; standalone native CLI binary (no external Node runtime)
  • Host: Codex; existing authorization worked for earlier synthesis chunks

Command (redacted)

bl speech synthesize --model cosyvoice-v3-flash --voice longanwen_v3 --rate 1.0 --pitch 1.0 --volume 50 --format wav --sample-rate 24000 --enable-ssml --text-file [REDACTED_LOCAL_SSML] --out [REDACTED_LOCAL_WAV] --output json --timeout 180

Expected

The command should complete successfully or return a structured timeout within the configured deadline, clearly marking an incomplete WAV as incomplete.

Observed

During a sequential Chinese TTS batch, one invocation remained running for at least 5m43s despite --timeout 180. Its WAV stopped at 540,995 bytes (11.2698 seconds of actual 24kHz mono PCM), far too short for the requested passage. The RIFF/data length fields were streaming placeholders. The parent was reading stdout/stderr via subprocess capture_output, so the output pipe was drained. No actionable error or request ID was available. The stuck child was then terminated; the incomplete audio and checkpoint were retained locally. The service-side billing/completion status is unknown; no claim is made about server cancellation. Re-generating that passage after checking the incomplete file completed normally. This report excludes user text, account information, credentials, internal addresses, and raw logs.

Frequency

Intermittent; a second invocation with the same configuration also stalled and was stopped by an external 240-second deadline. Its partial PCM lasted 13.3951 seconds; other chunks using identical parameters completed. A subsequent invocation completed after a long delay but within the external deadline. No extra paid diagnostic calls were made.

Suggested area to inspect

Whether the configured timeout covers the whole speech streaming/read loop, rather than only connection establishment, and how interrupted streaming output is finalized or flagged.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions