Skip to content

feat(regen)!: agent force-end-turn + flux expressivity; remove stale tts symbols - #100

Merged
GregHolmes merged 8 commits into
mainfrom
gh/sdk-gen-2026-09-07
Sep 11, 2026
Merged

GregHolmes merged 8 commits into
mainfrom
gh/sdk-gen-2026-09-07

Conversation

@GregHolmes

@GregHolmes GregHolmes commented Sep 7, 2026 •

Copy link
Copy Markdown
Collaborator

Fern SDK regeneration for Java SDK 0.9.0, targeting the pre-1.0 breaking 0.10.0 release.

Breaking changes

  • Remove the public SpeakV2Speed type. V2ConnectOptions#speed remains numeric (Double), so configure speed directly, for example .speed(1.05).
  • Remove AURA2PERSEO_IT from SpeakV1Model and AudioGenerateRequestModel, including Visitor.visitAura2PerseoIt(). The model was never served by the API (it returns 400) and was removed from the spec.

See the v0.9 to v0.10 migration guide for update steps.

Additions

  • Agent ForceEndTurn typed message and sendForceEndTurn(...) helper.
  • Flux Speak V2 expressivity plus the generated speed range of 0.5 to 1.5 in 0.05 increments.
  • Agent WebSocket ForceEndTurn wire coverage, asserting the exact {"type":"ForceEndTurn"} control frame.

Production availability

Flux TTS speeds range from 0.5 to 1.5 in 0.05 increments.

Listen V2 ForceEndTurn requires server-side enablement. Deployments without the feature return UNPARSABLE_CLIENT_MESSAGE and close the connection.

Patch reconciliation

  • Reapply SDK header/version, reconnect, streaming query serialization, forward compatibility, provider-union, and fieldless equals/hashCode fixes.
  • Add the fieldless equals/hashCode fix for AgentV1ForceEndTurn.
  • Drop the obsolete V2ConnectOptions compatibility patch because Fern now generates numeric speed correctly.
  • Remove all temporary .bak files.

Generator provenance

originGitCommitIsDirty is expected: Fern ran after an uncommitted fern/generators.yml change that only selected this Java output branch and push mode. No API definition or Java code-generation setting was uncommitted.

Validation

  • ./gradlew spotlessCheck test compileExamples
  • ./gradlew --rerun-tasks spotlessCheck test compileExamples

BREAKING CHANGE: SpeakV2Speed, SpeakV1Model.AURA2PERSEO_IT, AudioGenerateRequestModel.AURA2PERSEO_IT, and the corresponding Visitor.visitAura2PerseoIt() methods are removed. Use numeric V2ConnectOptions#speed and a supported Speak V1 model. See docs/Migrating-v0.9-to-v0.10.md.

@GregHolmes GregHolmes changed the title chore: SDK regeneration 2026-09-07 feat(regen)!: agent force-end-turn + flux expressivity; remove stale tts symbols Sep 7, 2026

@dg-coreylweathers dg-coreylweathers left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

DevRel review — changes requested. The code is correct and the real gate is green; the requests are about what this release promises developers.

What this PR does
Regenerates the Java SDK from the 2026-09-04 API spec with the same Fern generator (fern-java-sdk 4.18.0). Three things arrive: a sendForceEndTurn(...) helper that tells a Voice Agent session to end the user's turn now, an expressivity setting (-2 calm to 2 animated) on the agent's Deepgram speak provider for Flux voices, and Flux speed documented as 0.5–1.5 instead of 0.85–1.15. Two things leave: the SpeakV2Speed enum class (speed is a plain number now, so .speed(1.05) keeps compiling) and the AURA2PERSEO_IT voice constant in both Speak V1 model enums. Every hand-patch from earlier cycles is back in place.

What I checked

  • Does it build and pass? Yes — spotlessCheck test compileExamples in Docker (Temurin 17): 125 tests, 0 failures.
  • Did every hand-patch survive? Yes — reconnect, array query params, the additionalProperty escape hatch (wire test proves no_delay=true reaches the upgrade URL on all four streaming clients), forward-compat no-op, provider-union default, fieldless hashCode on 13 types, FLUX_RENEE_EN; .fernignore covers all 25 files; no .bak left.
  • Is the ForceEndTurn frame right? Yes — a wire test asserts the client sends exactly {"type":"ForceEndTurn"}.
  • Is the speed range true on production today? Not consistently — across two live probe rounds against /v2/speak?model=flux-kelsey-en, 0.85 always works; 0.5, 0.7, 1.5 were rejected with 'speed' must be between 0.85 and 1.15. on some requests and returned audio on others; 1.55 gets the new-range message. The fleet is mid-rollout. The README (README.md:341) and examples/speak/StreamingTtsV2.java:150 still say 0.85–1.15 while the new javadoc says 0.5–1.5.
  • Is aura-2-perseo-it really gone? Yes — the live API returns 400 No such model/version combination found for it (a real Italian voice, aura-2-livia-it, returns audio).
  • Are the breaking changes real and declared? Yes — both symbols shipped in 0.9.0 and the footer names them; feat(regen)! yields 0.10.0. JS #546 keeps both as deprecated aliases, so the two SDKs currently tell different stories.

What to fix / what still needs you

  1. (blocking) Speed range: please check with the Flux TTS team on when 0.5–1.5 is enforced fleet-wide (REST and mid-stream Configure). Until they confirm, either hold, or add a caveat to README and the PR body: "The spec documents 0.5–1.5; as of 2026-09-07 production accepts values outside 0.85–1.15 only intermittently — stay inside 0.85–1.15 until the wider range is announced live." Either way the README/example and javadoc need to agree.
  2. Footer wording for aura-2-perseo-it: state the reason — "never served by the API (returns 400) and removed from the spec" — and use the same sentence in the JS and Python changelogs; note that JS keeps a deprecated alias.
  3. Add docs/Migrating-v0.9-to-v0.10.md (include the removed Visitor.visitAura2PerseoIt() interface method) and link it from the footer, as 0.8.0 and 0.9.0 did.
  4. ClientOptions.java:53,56 say 0.9.1, a version that never existed. Unless there is a reason for stamping it, set back to 0.9.0; release-please bumps it to 0.10.0 at release (as #96 did).
  5. Restore the one-line // Manual patch — see .fernignore markers on the 21 patched files that lost them.

Also open: is Agent ForceEndTurn enabled on production everywhere? Same question as on the Python PR; one answer fixes the docstrings in all three SDKs. And originGitCommitIsDirty: true — what was uncommitted at 3121f56?

Nits and the full evidence table are in my review notes; happy to share.

@GregHolmes

Copy link
Copy Markdown
Collaborator Author

Addressed the review feedback in 8fcb16c. @dg-coreylweathers, could you please re-review when you have a chance?

@GregHolmes

Copy link
Copy Markdown
Collaborator Author

@dg-coreylweathers Follow-up on the Flux speed rollout: production verification on 2026-09-10 accepted all four boundary values: 0.5, 0.85, 1.15, and 1.5.

  • JS SDK: batch REST synthesis returned non-empty audio for each value; Speak V2 WebSocket Configure returned ConfigureSuccess and echoed the exact requested applied.speed for each.
  • Python SDK: the same REST and WebSocket matrix passed.
  • Java SDK: the four-value Speak V2 WebSocket Configure matrix passed with exact echoed values; its existing live streaming-audio integration also passed at 1.05. I did not run a separate Java REST boundary matrix.

No SPEED_OUT_OF_RANGE responses occurred at either new boundary. This resolves the prior fleet-rollout acceptance concern. A separate repeated batch-audio duration probe showed substantial variation at 0.5, so I would not use these results alone to claim strict rate-fidelity at the slow endpoint.

@GregHolmes

Copy link
Copy Markdown
Collaborator Author

@dg-coreylweathers, the requested Java updates are now in 2792356:

  • Replaced the stale speed rollout caveat with the 0.5 to 1.5 range in the README, Speak V2 example, migration guide, and PR body.
  • Added Gradle and Maven 0.10.0 snippets, corrected the removed-symbol explanation, and documented the SpeakV1Model.valueOf("aura-2-perseo-it") HTTP 400 behavior.
  • Added agent-provider expressivity(1) serialization coverage.
  • Added the migration-guide sentence to the BREAKING CHANGE footer.

./gradlew spotlessCheck test compileExamples passes. I left Agent ForceEndTurn Javadoc unchanged pending confirmation of its production enablement. Could you please re-review?

@dg-coreylweathers dg-coreylweathers left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

DevRel re-review — approved. Commit 2792356 resolves everything requested on 2026-09-11; the code surface is unchanged and nothing blocks merge.

What I checked

  • Does it build and pass? Yes — spotlessCheck test compileExamples in Temurin 17 (Docker): 126 tests, 0 failures, including the new RegenTypesTest$AgentSpeakProvider case. CI on 2792356 is green on all 8 checks.
  • Is any stale speed caveat left? No — a repo-wide grep for "intermittent", "announced live", "as of 2026-09", and "0.85" finds nothing outside CHANGELOG.md and the historical v0.8→v0.9 guide. README, example, guide, and PR body all say 0.5–1.5 in 0.05 steps, which production returned on 32/32 boundary requests today (REST and mid-stream Configure).
  • Will the CHANGELOG link the migration guide? Yes — the BREAKING CHANGE: footer now ends "See docs/Migrating-v0.9-to-v0.10.md.", the same shape that produced the 0.9.0 link. Please keep the footer in the squash commit body at merge time.
  • Guide completeness: Gradle and Maven snippets at 0.10.0, corrected intro (only AURA2PERSEO_IT was "never served"; SpeakV2Speed was an SDK-side type), and the SpeakV1Model.valueOf("aura-2-perseo-it") HTTP 400 note are all present.
  • Sibling parity: JS #546 and Python #783 bodies no longer carry the caveat either.

Agent ForceEndTurn is deployed on production, so the Agent sendForceEndTurn javadoc is correct as written with no enablement note. One line to drop as a follow-up (or before merge if convenient): docs/Migrating-v0.9-to-v0.10.md "New features" says "ForceEndTurn still requires server-side enablement; deployments without it return UNPARSABLE_CLIENT_MESSAGE and close the connection." That describes Listen V2 from 0.9.0, not the Agent helper this release adds. Suggested replacement: "sendForceEndTurn(...) and Flux speak-provider expressivity are additive and available on production."

@GregHolmes
GregHolmes merged commit cbadf05 into main Sep 11, 2026
8 checks passed
GregHolmes added a commit that referenced this pull request Sep 14, 2026
🤖 I have created a release *beep* *boop*
---

##
[0.10.0](v0.9.0...v0.10.0)
(2026-09-11)

Agent force-end-turn support, Flux TTS expressivity controls, and a
wider Flux TTS speed range. This pre-1.0 minor release also removes
stale TTS symbols.

### ⚠ BREAKING CHANGES

* **Speak v2 (Flux TTS):** the `SpeakV2Speed` type and its named speed
constants are removed. `V2ConnectOptions#speed` remains a numeric
`Double`; configure speed directly, for example `.speed(1.05)`.
* **Speak v1 (Aura TTS):** remove `SpeakV1Model.AURA2PERSEO_IT`,
`AudioGenerateRequestModel.AURA2PERSEO_IT`, their `Value` enum entries,
and the associated `Visitor.visitAura2PerseoIt()` methods. The
`aura-2-perseo-it` model was never served by the API and is no longer in
the specification. Use a supported Speak v1 model instead. See the [v0.9
to v0.10 migration
guide](https://github.com/deepgram/deepgram-java-sdk/blob/main/docs/Migrating-v0.9-to-v0.10.md).

### Features

* **Agent:** add `AgentV1ForceEndTurn` and
`V1WebSocketClient#sendForceEndTurn(...)` to end the current turn on
demand. Agent Deepgram speak providers using Flux (`version: "v2"`) also
accept `expressivity`, a whole-number calm-to-animated setting from `-2`
to `2`; `0` is the default.
([#100](#100))
([cbadf05](cbadf05))
* **Speak v2 (Flux TTS):** batch requests and WebSocket connections
accept `expressivity` from `-2` to `2`. `speed` now accepts values from
`0.5` to `1.5` in `0.05` increments, instead of `0.85` to `1.15`.
([#100](#100))
([cbadf05](cbadf05))

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Greg Holmes <greg.holmes@deepgram.com>
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.

2 participants