Deepgram Flux TTS in EVA

.env.example

@@ -90,14 +90,17 @@ EVA_MODEL__STT_PARAMS='{"api_key": "your_cartesia_api_key", "model": "ink-2"}'
 # --- TTS (LLM and AudioLLM modes) ---
 #i TTS provider for the voice pipeline.
 #d enum
-#e cartesia,chatterbox,elevenlabs,gemini,kokoro,nvidia-baseten,openai,smallest,soniox,xtts
+#e cartesia,chatterbox,deepgram,deepgram-flux,elevenlabs,gemini,kokoro,nvidia-baseten,openai,smallest,soniox,xtts
 #x pipeline_mode=LLM,AudioLLM
 EVA_MODEL__TTS=cartesia
 
 #i TTS provider parameters. Must include "api_key" and "model". Use "urls" for round-robin load balancing, and "voice_id" to select a voice.
 #i Some providers also accept extra provider-specific tuning parameters here.
 #i Soniox example — omit "voice_id" to use pipecat's default voice ("Adrian"):
 #i EVA_MODEL__TTS_PARAMS='{"api_key": "your_soniox_api_key", "model": "tts-rt-v2", "voice": "Sarah"}'
+#i Deepgram Flux example (EVA_MODEL__TTS=deepgram-flux) — "voice" is the full flux-{voice}-{lang} model
+#i string (defaults to "flux-alexis-en" if omitted); "model" is not used for Flux:
+#i EVA_MODEL__TTS_PARAMS='{"api_key": "your_deepgram_api_key", "voice": "flux-haley-en"}'
 #d json_object
 #x pipeline_mode=LLM,AudioLLM
 EVA_MODEL__TTS_PARAMS='{"api_key": "your_cartesia_api_key", "model": "sonic"}'

pyproject.toml

@@ -21,7 +21,7 @@ license = "MIT"
 dependencies = [
     "pydantic>=2.0",
     "pydantic-settings>=2.0",
-    "pipecat-ai>=1.4.0",
+    "pipecat-ai>=1.6.0",
     "elevenlabs>=2.53.0",
     "openai>=2.36.0",
     "anthropic>=0.83.0",

src/eva/assistant/pipeline/services.py

@@ -21,6 +21,7 @@
 from pipecat.services.cartesia.tts import CartesiaTTSService
 from pipecat.services.cartesia.turns.stt import CartesiaTurnsSTTService
 from pipecat.services.deepgram.flux.stt import DeepgramFluxSTTService, DeepgramFluxSTTSettings
+from pipecat.services.deepgram.flux.tts import DeepgramFluxTTSService
 from pipecat.services.deepgram.stt import DeepgramSTTService
 from pipecat.services.deepgram.tts import DeepgramTTSService
 from pipecat.services.elevenlabs.stt import CommitStrategy, ElevenLabsRealtimeSTTService
@@ -375,7 +376,7 @@ def create_tts_service(
     Based on create_tts_service() from chatbot.py.
 
     Args:
-        model: TTS model identifier (cartesia, elevenlabs, openai, gemini)
+        model: TTS model identifier (cartesia, deepgram, deepgram-flux, elevenlabs, openai, gemini)
         params: Model-specific parameters (may include 'alias' key which is ignored here)
         language_code: Language code for speech synthesis
 
@@ -430,7 +431,17 @@ def create_tts_service(
         chatterbox_tts._settings.language = language_code
         return chatterbox_tts
 
-    elif model_lower == "deepgram":
+    elif model_lower.startswith("deepgram"):
+        if "flux" in model_lower:
+            # Flux TTS (/v2/speak) has no separate "model" field: the flux-{voice}-{lang}
+            # string IS the voice, and pipecat keeps Settings.model in sync with it.
+            voice = params.get("voice", "flux-alexis-en")
+            logger.info(f"Using Deepgram Flux TTS: {voice}")
+            return DeepgramFluxTTSService(
+                api_key=api_key,
+                sample_rate=SAMPLE_RATE,
+                settings=DeepgramFluxTTSService.Settings(voice=voice),
+            )
         logger.info(f"Using Deepgram TTS: {params['model']}")
         return DeepgramTTSService(
             api_key=api_key,
@@ -646,7 +657,7 @@ def create_tts_service(
 
     else:
         raise ValueError(
-            f"Unknown TTS model: {model}. Available: cartesia, chatterbox, deepgram, elevenlabs, gemini, kokoro, nvidia-baseten, openai, smallest, soniox, voxtral, xai, xtts"
+            f"Unknown TTS model: {model}. Available: cartesia, chatterbox, deepgram, deepgram-flux, elevenlabs, gemini, kokoro, nvidia-baseten, openai, smallest, soniox, voxtral, xai, xtts"
         )
 
 

tests/unit/assistant/test_services.py

@@ -321,6 +321,21 @@ def test_soniox_defaults_voice_when_unset(self):
         svc = create_tts_service("soniox", params={"api_key": "k", "model": "tts-rt-v1"})
         assert svc._settings.voice == "Adrian"
 
+    def test_deepgram_returns_deepgram_service(self):
+        svc = create_tts_service("deepgram", params={"api_key": "k", "model": "aura-2-helena-en"})
+        assert "Deepgram" in type(svc).__name__
+        assert "Flux" not in type(svc).__name__
+        assert svc._settings.voice == "aura-2-helena-en"
+
+    def test_deepgram_flux_returns_flux_variant(self):
+        svc = create_tts_service("deepgram-flux", params={"api_key": "k", "voice": "flux-haley-en"})
+        assert "Flux" in type(svc).__name__
+        assert svc._settings.voice == "flux-haley-en"
+
+    def test_deepgram_flux_defaults_voice_when_unset(self):
+        svc = create_tts_service("deepgram-flux", params={"api_key": "k"})
+        assert svc._settings.voice == "flux-alexis-en"
+
     def test_openai_respects_voice_param(self):
         svc = create_tts_service("openai", params={"api_key": "k", "model": "tts-1", "voice": "nova"})
         assert svc._settings.voice == "nova"