Skip to content

feat(java): Add a standalone Java ASR SDK POC - #1094

Closed
Mason Chen (jiec-msft) wants to merge 5 commits into
microsoft:mainfrom
jiec-msft:mason/java-asr-sdk
Closed

Mason Chen (jiec-msft) wants to merge 5 commits into
microsoft:mainfrom
jiec-msft:mason/java-asr-sdk

Conversation

@jiec-msft

Copy link
Copy Markdown
Member

Add a standalone Java 17/JNA ASR POC with pinned native/model metadata, bounded WAV/PCM input, cancellation and UTF-8 JSONL.

The fixed JAR passed five-target native smoke on ten public utterances; evidence and limits keep binary and metadata provenance separate.

Draft only: incomplete per-file exports, old-JBR compatibility, offline behavior and redistribution remain review limits, not product readiness.

Bind the pinned public Runtime 2.0.1 API 1 ABI through JNA, with explicit preparation, bounded PCM streaming and deterministic ownership. Include CLI schemas, platform locks and public speech smoke evidence. Document the original JBR 21 CRT incompatibility and separately supported JBR 25 experiment without claiming strict offline operation or cross-platform validation.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Write complete JSONL events as explicit UTF-8 bytes. Keep cancellation and terminal publication in one native-independent completion state, rechecking cancellation atomically before exposing the result. Add a real subprocess encoding regression and a latch-controlled regression at the native decoder/publication seam. Preserve prior native evidence as baseline-only pending regression of the rebuilt JAR.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Preserve the original Windows model lock and qualified d094 Java binary. Add a reviewed metadata sidecar for Windows x64/ARM64 and the observed Linux x64 generated marker, with complete per-target inventory hashes and totals. Keep unobserved targets fail-closed, share all fifteen common pins, and document separate evaluator metadata pinning with pure metadata regressions.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Promote only the Linux ARM64 and macOS ARM64 sidecar entries using their separate public diagnostic provenance and exact raw inventory tuples. Keep inventory-only observations distinct from ASR qualification, retain all common and Windows pins, and extend pure metadata tests across all five selections and simulated unobserved failures. The schema, selector API and qualified d094 binary are unchanged.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Link the immutable second hosted matrix report, exact binary and metadata pins, actual lifecycle/sample counts and bounded performance observations. Preserve historical Windows/JBR/ABI evidence and clearly disclose the omitted success-inventory rows, unknown transfer metrics and POC-only scope. Documentation only; no implementation, lock, test, workflow or evaluator history changes.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@vercel

vercel Bot commented Sep 9, 2026

Copy link
Copy Markdown

Mason Chen (@jiec-msft) is attempting to deploy a commit to the MSFT-AIP Team on Vercel.

A member of the Team first needs to authorize it.

@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

@jiec-msft

Copy link
Copy Markdown
Member Author

Replaced by #1100. The original mason/java-asr-sdk branch remains unchanged for the existing POC.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant