fix(backend): stop expiring speech profiles in has_profile check (#5128)

[kodjima33]

Bug (#5128, Bug 1)

Users who already have a speech profile are re-prompted to "Teach Omi your voice". Live on prod for every user whose profile is older than 90 days (i.e. most long-time users).

Root cause

Commit 34be170 (#3891, Dec 24 2025) added a 90-day expiry to the /v3/speech-profile endpoint only:

return {'has_profile': get_user_has_speech_profile(uid, max_age_days=90)}

Nothing else honors that expiry:

• routers/transcribe.py:764 enables speaker identification via get_user_has_speech_profile(uid) — no age limit
• routers/transcribe.py:1833 downloads the profile for embedding extraction regardless of age
• get_profile_audio_if_exists() has no age limit

So a >90-day profile is actively used for diarization while the app is told has_profile: false and shows the re-teach banner every launch. The expiry never had the (presumably intended) effect of refreshing what the pipeline uses — its only production effect is the spurious nag.

Fix

Remove the age cutoff from get_user_has_speech_profile so the banner endpoint agrees with the listen pipeline: an existing profile is a profile. Zero pipeline behavior change. Bonus: drops the extra per-request GCS metadata round-trip (blob.reload()) the age check required.

max_age_days had no other callers, so the dead parameter is removed too.

Tests

tests/unit/test_speech_profile_existence.py (registered in test.sh): existence semantics (old profile counts, missing blob/bucket don't), no blob.reload() metadata fetch, signature guard (uid only), and a structural guard that the router passes no age cutoff. All assertions verified locally on Python 3.11 against the real modules (pytest not installed locally; CI runs the suite). black clean, lint_async_blockers clean.

Note: Bug 2 in #5128 (recording UX glitches) is speculative/multi-cause and is not addressed here.

🤖 automated by hourly watchdog; opened for review, not merged.

Fixes #5128

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

This PR fixes a regression where users with speech profiles older than 90 days were incorrectly shown the "Teach Omi your voice" re-enrollment banner on every launch, because the /v3/speech-profile endpoint was filtering profiles by age while the transcription pipeline used them without any age restriction.

• utils/other/storage.py: Removes the max_age_days parameter and inline age-check logic from get_user_has_speech_profile, collapsing it to a single blob.exists() call. This also eliminates the extra blob.reload() GCS metadata round-trip that the expiry check required.
• routers/speech_profile.py: Drops the max_age_days=90 argument so the /v3/speech-profile endpoint now agrees with the transcription pipeline on what constitutes a valid profile.
• tests/unit/test_speech_profile_existence.py + test.sh: Adds a new test suite covering profile existence semantics, the absence of blob.reload(), the function signature (no age parameter), and a structural guard that the router does not reintroduce an age cutoff.

Confidence Score: 4/5

Safe to merge — the change removes a one-line age filter that was never applied by the transcription pipeline, restoring consistency between the API and the actual diarization behavior.

The core logic change is minimal and verifiably correct: both call sites in transcribe.py and speech_profile.py already passed no age limit, so collapsing to a plain blob.exists() call only removes the divergence. The one point of minor attention is the test test_endpoint_does_not_pass_age_cutoff, which derives the router path from storage_mod.file at runtime — correct given the backend/utils/other/storage.py nesting, but would silently break if the module path ever changes.

No files require special attention. The structural test in test_speech_profile_existence.py warrants a glance if the directory layout changes in the future.

Important Files Changed

Filename	Overview
backend/utils/other/storage.py	get_user_has_speech_profile simplified to a single blob.exists() call; max_age_days parameter and blob.reload() metadata fetch removed; all other callers used no age limit anyway
backend/routers/speech_profile.py	Endpoint call updated to drop max_age_days=90; now consistent with the transcribe.py pipeline call at line 764
backend/tests/unit/test_speech_profile_existence.py	New unit tests covering existence semantics, missing bucket/blob, signature guard, and a filesystem-based structural guard that the router omits max_age_days; the filesystem read in test_endpoint_does_not_pass_age_cutoff is slightly brittle but functionally correct
backend/test.sh	New test module registered in the test suite alongside existing unit tests

Sequence Diagram

sequenceDiagram
    participant App as Mobile App
    participant EP as /v3/speech-profile
    participant Storage as utils/other/storage.py
    participant GCS as Google Cloud Storage
    participant TP as Transcription Pipeline

    Note over App,TP: Before fix (regression)
    App->>EP: GET /v3/speech-profile
    EP->>Storage: "get_user_has_speech_profile(uid, max_age_days=90)"
    Storage->>GCS: blob.exists()
    Storage->>GCS: blob.reload() [fetch metadata]
    Storage-->>EP: "False (profile >90 days old)"
    EP-->>App: "{has_profile: false}"
    App-->>App: Show re-enroll banner

    TP->>Storage: get_user_has_speech_profile(uid)
    Storage->>GCS: blob.exists()
    Storage-->>TP: True
    TP-->>TP: Uses profile for diarization

    Note over App,TP: After fix (this PR)
    App->>EP: GET /v3/speech-profile
    EP->>Storage: get_user_has_speech_profile(uid)
    Storage->>GCS: blob.exists()
    Storage-->>EP: True
    EP-->>App: "{has_profile: true}"

    TP->>Storage: get_user_has_speech_profile(uid)
    Storage->>GCS: blob.exists()
    Storage-->>TP: True
    TP-->>TP: Uses profile for diarization

Loading

_{Reviews (1): Last reviewed commit: "fix(backend): stop expiring speech profi..." | Re-trigger Greptile}