feat(speak): default to Flux TTS (flux-alexis-en) instead of Aura 2

[GregHolmes]

What

Changes the default dg speak model from aura-2-asteria-en to flux-alexis-en, so a bare dg speak "..." now uses Flux TTS (Speak v2, WebSocket streaming) out of the box. Aura remains fully available via -m aura-*.

Follow-on to the Speak v2 / Flux TTS implementation (#86). Asana: DX CLI task.

Why

Flux TTS is our latest, most natural TTS. Making it the default puts our best voice front-and-center in the CLI.

How it works

Routing is unchanged and purely prefix-based (model.lower().startswith("flux")): flux-* goes to Speak v2 streaming, everything else goes to Speak v1 (Aura, REST batch). Picking the voice is picking the API path -- there is no separate mode flag.

Behavior change

Because the default now takes the v2 streaming path (raw linear16 auto-wrapped in WAV), containerized formats like mp3 now require an explicit -m aura-*:

• dg speak "hi" -o out.wav -> Flux, WAV (new default)
• dg speak "hi" -o out.mp3 --encoding mp3 -> fails with a clear error + non-zero exit (was previously fine under the Aura default)
• dg speak "hi" -o out.mp3 -m aura-2-asteria-en -> Aura v1 REST, mp3 (opt back in)

This is worth a line in the release notes.

Not changed

Core client speak_text() default stays aura-2-asteria-en -- it is the Aura-only v1 REST path (speak.v1.audio.generate); Flux goes through speak_text_stream().

Testing

• Unit: 20/20 pass, incl. new test_handle_default_model_routes_to_flux (asserts no -m routes to Flux v2, not Aura v1).
• Manual (branch dg + released SDK 7.5.0, against prod):
  • bare dg speak -> Flux streaming, plays correctly
  • -m aura-2-asteria-en -> Aura v1 REST, plays correctly
  • default + --encoding mp3 -> fails clean, exit 2, no file written

Open question

Flux TTS is currently EA with a single voice (flux-alexis-en); more land at GA (~next month). Decision needed: merge now and ship Flux-by-default against the single EA voice, or hold this until GA when the full roster is available.

[dg-coreylweathers]

Reviewed — code and docs are correct, tests pass locally (speak 20/20, including the new default-routes-to-Flux test), routing and the Aura REST path are unchanged, and client.speak_text() correctly stays aura-2-asteria-en. Approving the content; EA-single-voice ship is a go on our side.

One thing to fix before merge, and one follow-up.

⚠️ Before merge — signal the breaking change to release-please

The change is an intentional break (bare dg speak -o x.mp3 now fails; default output is WAV/linear16 instead of MP3), but the commit/PR title is a plain feat(speak):. Release-please generates the changelog and version bump from that prefix, so as-is this ships as a patch under "Features" with the break invisible. Since we squash-merge (PR title + body → commit), please:

• Title → feat(speak)!: default to Flux TTS (flux-alexis-en) instead of Aura 2
• Add to the body:

  BREAKING CHANGE: bare `dg speak` now defaults to Flux TTS (Speak v2 streaming),
  emitting WAV/linear16 @ 24 kHz instead of MP3; `--container` is ignored on the
  default path. To restore the previous Aura/MP3 behavior, pass `-m aura-2-asteria-en`.
  

(No code change needed — this is purely the release signal. The spec's "add a CHANGELOG entry" step doesn't apply here since release-please auto-generates it.)

Follow-up (non-blocking)

packages/deepctl-core/src/deepctl_core/skill_generator.py:420-443 wasn't updated — the generated agent skill still opens with "…using Deepgram's Aura voices" and lists Aura first, Flux last, which contradicts the new Flux-by-default CLI. Worth reordering to lead with Flux, either here or as a fast follow.

Notes

• Default + --encoding mp3 fails cleanly before any write (no empty/header-only file) — verified.
• The 5 test_base.py::test_output_result_* failures pre-exist on main; unrelated to this PR.

[dg-coreylweathers]

Approved. Fix the feat!: title + BREAKING CHANGE footer before merge; EA-single-voice ship is a go.