Remove Cartesia provider — keep AWS Polly + Microsoft Azure only#80
Merged
Conversation
|
Warning Rate limit exceeded
You’ve run out of usage credits. Purchase more in the billing tab. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (3)
WalkthroughPremium TTS 提供元から Cartesia サポートを削除し、Polly（AWS）と Azure（Microsoft）のみに統一。TTSキャッシュキー契約を v5→v6（Worker は v2→v3）に更新して model パラメータを廃止。iOS と Worker のストリーミング経路を生 PCM 前提で整理し、LapAnnouncer のプリウォーム並行制御・キャンセル再開ロジックを追加。 ChangesCartesia 削除と Premium TTS Polly/Azure 統一
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
app/HDZap/Models/LapAnnouncer.swift (1)
1025-1034:⚠️ Potential issue | 🟠 Major | ⚡ Quick winPremium を外したときも既存 prewarm を止めてください。
Line 1026 の
guardが先にreturnするので、設定変更で system に戻したり voice を未選択にした直後でも、前回のcurrentPrewarmTaskは走り続けます。不要な prefetch が残ると upstream への負荷と 429 リスクをまた増やすので、キャンセルはguardより前で必ず実行した方がいいです。💡 修正案
func prewarmFixedPhrases() { + currentPrewarmTask?.cancel() + currentPrewarmTask = nil guard let voice = currentPremiumVoiceIfActive() else { return } let phrases = fixedPrewarmPhrases(for: currentLanguage()) let synth = premiumSynth let lang = voice.lang - currentPrewarmTask?.cancel() - currentPrewarmTask = Task.detached { `@MainActor` in + currentPrewarmTask = Task.detached { `@MainActor` in🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@app/HDZap/Models/LapAnnouncer.swift` around lines 1025 - 1034, In prewarmFixedPhrases(), always cancel any existing currentPrewarmTask before the early exit: move the currentPrewarmTask?.cancel() call to the top of the function (before the guard let voice = currentPremiumVoiceIfActive() else { return }) so switching off premium or deselecting voice stops any running prewarm; keep the rest of the logic (computing phrases, synth, lang, and creating Task.detached) unchanged and ensure currentPrewarmTask is then set to the new Task only when a new prewarm starts.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@app/HDZap/Models/LapAnnouncer.swift`:
- Around line 355-367: premiunUtteranceEnded currently always calls
prewarmFixedPhrases after utteranceDidEnd/deactivateSession, which can restart
prewarm while a replacement utterance is still playing; modify
premiumUtteranceEnded to only call prewarmFixedPhrases when there are no active
utterances (check inflightUtteranceCount == 0) after invoking utteranceDidEnd(),
keeping deactivateSession() as-is; reference premiumUtteranceEnded,
utteranceDidEnd, prewarmFixedPhrases, deactivateSession, and
inflightUtteranceCount when making the change.
In `@app/HDZap/Models/Speech/PremiumSpeechSynthesizer.swift`:
- Around line 578-590: The retry path swallows Task cancellation because `try?
await Task.sleep(...)` ignores CancellationError and causes a retry even after
prewarm was cancelled; in PremiumSpeechSynthesizer's prewarm code replace the
`try? await Task.sleep(nanoseconds: 1_000_000_000)` with a
cancellation-respecting approach (e.g. use `try await Task.sleep(...)` and let
CancellationError propagate, or explicitly check `Task.isCancelled` and abort
before re-sending the request) so the subsequent `(data, response) = try await
URLSession.shared.data(for: req)` is not executed when the task has been
cancelled.
---
Outside diff comments:
In `@app/HDZap/Models/LapAnnouncer.swift`:
- Around line 1025-1034: In prewarmFixedPhrases(), always cancel any existing
currentPrewarmTask before the early exit: move the currentPrewarmTask?.cancel()
call to the top of the function (before the guard let voice =
currentPremiumVoiceIfActive() else { return }) so switching off premium or
deselecting voice stops any running prewarm; keep the rest of the logic
(computing phrases, synth, lang, and creating Task.detached) unchanged and
ensure currentPrewarmTask is then set to the new Task only when a new prewarm
starts.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 5245d31b-7132-4dfb-b85b-56f7e20b7927
⛔ Files ignored due to path filters (1)
workers/hdzap-premium/package-lock.jsonis excluded by!**/package-lock.json
📒 Files selected for processing (10)
app/HDZap.xcodeproj/project.pbxprojapp/HDZap/Models/LapAnnouncer.swiftapp/HDZap/Models/Speech/PremiumSpeechSynthesizer.swiftapp/HDZap/Models/Speech/TTSCache.swiftapp/HDZap/Resources/StoreKit/HDZapPremium.storekitapp/HDZap/Views/Settings/AudioSettingsView.swiftapp/HDZap/Views/Settings/PaywallView.swiftapp/HDZap/Views/Settings/PremiumVoicePickerView.swiftapp/project.ymlworkers/hdzap-premium/src/index.ts
Saqoosha
force-pushed
the
remove-cartesia-provider
branch
2 times, most recently
from
78ff8aa to
f0cf7ba
Compare
- PremiumVoiceProvider enum loses the .cartesia case. PremiumVoiceCatalog drops 25 Cartesia voices (22 JA + 3 EN); Polly (14) + Azure (16) = 30 voices total, plenty for race-announcer / friendly-narrator personas across US/UK/AU accents
- Worker drops the proxyCartesia function, the CARTESIA_API_KEY env binding, the ALLOWED_CARTESIA_MODELS allow-list, the cartesia branch in the /tts handler, and the cartesia / SSE special-cases in contentTypeFor / sampleRateFor / responseHeadersFor. The Cartesia model field is removed from buildCacheKey; R2 key prefix bumps v2 -> v3 to invalidate the old entries
- iOS drops parseSSE + handleEventJSON (SSE was Cartesia-only — Polly + Azure stream raw PCM directly). prefetch + sendAndStream collapse to a single Polly/Azure raw-PCM path. accumulatedPCM, currentSampleRate, and the resampler logic stay (Polly is still 16 kHz, needs upsample). TTSCache key drops the trailing model segment and bumps v5 -> v6
- All Cartesia comments / labels / hint strings removed from PaywallView, PremiumVoicePickerView, AudioSettingsView, LapAnnouncer, and the HDZapPremium.storekit product descriptions ('30+ voices across AWS Polly and Microsoft Azure'). Sample voice teaser on paywall is now 2-row (Polly + Azure) instead of 3-row
- Motivation: Cartesia's per-IP rate limit on parallel SSE bursts kept dropping prewarm prefetches (observed 8 of 14 phrases 429'd at countdownStartSeconds=15 with the 3-concurrent cap from PR #79). Polly + Azure are both more permissive on TPS, so removing Cartesia eliminates the rate-limit class of bugs entirely while retaining a strong race-announcer voice catalog
- Code simplification: ~450 lines removed across Worker + iOS, no SSE parsing, no model-field special-cases, no Cartesia-specific UI disclaimers. Lower-surface-area Premium TTS for the remaining roadmap work
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Saqoosha
force-pushed
the
remove-cartesia-provider
branch
from
f0cf7ba to
25234da
Compare
…lyphase tail Polly 16 kHz countdown utterances were truncating ~10-20 ms at the end because the one-shot resampler input block returned .noDataNow instead of .endOfStream. The polyphase upsampler's FIR-filter tail stayed buffered inside AVAudioConverter and never reached the output buffer. Azure (24 kHz native) bypasses the converter so was unaffected. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- PremiumVoicePickerView: selecting a premium voice now also flips `ttsEngine` to "premium" (both on direct tap when entitled and on post-purchase auto-commit). Previously the selectedId was written but the router kept routing through System voice, so the picker looked like it did nothing. - AudioSettingsView: the Voice section's Reset button no longer touches Announcement-section keys (master toggle, announce-best, countdown enable/start). It now also resets ttsEngine and the premium voice / rate / pitch so a single tap returns the entire Voice section to defaults, while leaving the Announcement section untouched. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Saqoosha
force-pushed
the
remove-cartesia-provider
branch
from
25234da to
01ba91c
Compare
…rsions Two defensive hardenings in `buildOverlapBuffer` that protect against the exact silent-failure mode the prior `.endOfStream` fix targeted, in case AVAudioConverter behaves at the edges of its documented contract. - Raise `outputCapacity` headroom from +64 to +1024 frames. The FIR group delay for 16 kHz → 24 kHz upsample is ~300-600 output samples; +64 (~2.7 ms) leaves no margin if Apple's converter ever needs slightly more room after `.endOfStream` than the prior frame ratio suggested. +1024 (~43 ms) costs ~2 KB per utterance and gives a 2× safety factor. - Throw on `outputBuffer.frameLength == 0`. `convert()` can return a non-error status while still producing zero frames; scheduling such a buffer on AVAudioPlayerNode is a silent no-op (no completion callback path that signals failure). The throw lets `speakOverlap`'s catch set `lastError` so the missing audio is visible in the Settings banner / dev panel instead of an unexplained silent countdown beat. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Saqoosha
added a commit
that referenced
this pull request
May 22, 2026
…ning Build-number-only bump — MARKETING_VERSION stays at 1.1.0 so the build ships to existing 1.1.0 beta testers via Apple's fast-track build review, without re-running beta-review approval. Includes PR #80 in develop: Cartesia provider removed, Polly countdown tail truncation fixed (.endOfStream flush), voice picker engine flip on selection, Voice section Reset scope narrowed, AVAudioConverter defensive zero-frame guard + wider FIR tail headroom. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
PremiumVoiceProviderloses the.cartesiacase.PremiumVoiceCatalogdrops 25 Cartesia voices (22 JA + 3 EN). Polly (14) + Azure (16) = 30 voices still covers US/UK/AU + JA/EN with race-announcer and friendly-narrator personas.proxyCartesiadeleted,CARTESIA_API_KEYbinding removed,ALLOWED_CARTESIA_MODELSallow-list gone, cartesia branches removed fromcontentTypeFor/sampleRateFor/responseHeadersFor.buildCacheKeydrops themodelfield; R2 key prefix bumpsv2 → v3.parseSSE+handleEventJSONdeleted (SSE was Cartesia-only — Polly + Azure stream raw PCM directly).prefetch+sendAndStreamcollapse to a single Polly/Azure raw-PCM path.TTSCache.keydrops themodelfield; local cache prefix bumpsv5 → v6.PaywallView,PremiumVoicePickerView,AudioSettingsView,LapAnnouncer, and theHDZapPremium.storekitproduct descriptions ("30+ voices across AWS Polly and Microsoft Azure"). Paywall sample teaser is now 2-row (Polly + Azure) instead of 3-row.Why
TestFlight Build 17 tester report: Cartesia's per-IP rate limit on parallel SSE bursts kept dropping prewarm prefetches — even with the 3-concurrent cap added in PR #79, 8 of 14 phrases returned 429 at
countdownStartSeconds=15and Start audio failed to play because the user-visible request hit the rate-limit ceiling. Polly + Azure are both materially more permissive on TPS, so removing Cartesia eliminates the entire rate-limit class of bugs while retaining a strong race-announcer voice catalogue.Bonus: ~450 lines removed across Worker + iOS, no more SSE parsing, no model-field special-cases, no Cartesia-specific UI disclaimers. Lower-surface-area Premium TTS for the remaining roadmap work.
Test plan
.cartesiareferences) — already done locally,** BUILD SUCCEEDED **.wrangler deploy) and confirm a/ttsPOST withprovider=pollyand one withprovider=azureeach return audio. A POST withprovider=cartesiashould now returnbad-provider(400).premiumLapVoiceIdUUID), the voice falls back to the System path becausecurrentPremiumVoiceIfActive()returns nil (voice ID no longer in catalog). Re-pick a Polly or Azure voice.Summary by CodeRabbit
Chores
Improvements
UI Updates