mARTin-B78/tts-voice-creator-clone-and-design-2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update engine RAM metric chips to measured values from docker stats

Browse Source 
Parakeet RNNT NIM measured at 550–605 MB (unified RAM), Magpie TTS at
982 MB, and Qwen3-TTS model at 4.3 GB on disk — replacing the placeholder
estimates that were too high across the board.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
mARTin-B78 2026-05-27 23:51:15 +02:00
parent 6a78be7a28
commit 49e48eb4e3
1 changed files with 9 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
server.py
View File

@ -1820,7 +1820,7 @@ _STT_BACKEND_METRICS: dict[str, dict] = {
"faster_whisper": {"speed": "~70× RT · GPU", "latency": "0.5–2 s", "quality": "large-v3", "ram": "1.5 GB VRAM"}, "faster_whisper": {"speed": "~70× RT · GPU", "latency": "0.5–2 s", "quality": "large-v3", "ram": "1.5 GB VRAM"},
"whisper_cpp": {"speed": "~8–15× RT · CPU", "latency": "1–5 s", "quality": "large-v3 Q5", "ram": "~1 GB RAM"}, "whisper_cpp": {"speed": "~8–15× RT · CPU", "latency": "1–5 s", "quality": "large-v3 Q5", "ram": "~1 GB RAM"},
"groq_whisper": {"speed": "fastest cloud", "latency": "0.5–1 s", "quality": "Whisper Turbo", "ram": "cloud · 0"}, "groq_whisper": {"speed": "fastest cloud", "latency": "0.5–1 s", "quality": "Whisper Turbo", "ram": "cloud · 0"},
"nvidia_parakeet":{"speed": "~200× RT · GPU", "latency": "<0.3 s", "quality": "Parakeet-TDT", "ram": "2 GB VRAM"}, "nvidia_parakeet":{"speed": "~200× RT · GPU", "latency": "<0.3 s", "quality": "Parakeet-TDT", "ram": "~1 GB"},
"nvidia_router": {"speed": "GPU routed", "latency": "~0.5 s", "quality": "varies", "ram": "varies"}, "nvidia_router": {"speed": "GPU routed", "latency": "~0.5 s", "quality": "varies", "ram": "varies"},
} }
@ -3213,7 +3213,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Weak per-request style; instruct may be ignored.", "style": "Weak per-request style; instruct may be ignored.",
"best_for": "Known voices, multilingual cloning, benchmarks, and reliable speaker identity.", "best_for": "Known voices, multilingual cloning, benchmarks, and reliable speaker identity.",
"uses_wav": True, "style_aware": False, "true_streaming": False, "uses_wav": True, "style_aware": False, "true_streaming": False,
"speed": "~0.3× GPU", "latency": "1–3 s", "quality": "Premium clone", "ram": "6–8 GB VRAM", "speed": "~0.3× GPU", "latency": "1–3 s", "quality": "Premium clone", "ram": "~5 GB",
}, },
"voice_design": { "voice_design": {
"purpose": "Create or reuse prompt-designed voices from natural-language descriptions.", "purpose": "Create or reuse prompt-designed voices from natural-language descriptions.",
@ -3221,7 +3221,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Strong style and emotion control through instruct text.", "style": "Strong style and emotion control through instruct text.",
"best_for": "New characters, personas, dialogue, and designing reference WAVs to clone later.", "best_for": "New characters, personas, dialogue, and designing reference WAVs to clone later.",
"uses_wav": False, "style_aware": True, "true_streaming": False, "uses_wav": False, "style_aware": True, "true_streaming": False,
"speed": "~0.4× GPU", "latency": "1–3 s", "quality": "Premium", "ram": "6–8 GB VRAM", "speed": "~0.4× GPU", "latency": "1–3 s", "quality": "Premium", "ram": "~5 GB",
}, },
"customvoice": { "customvoice": {
"purpose": "Generate speech with the CustomVoice model voices.", "purpose": "Generate speech with the CustomVoice model voices.",
@ -3229,7 +3229,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Good per-request style and emotion control.", "style": "Good per-request style and emotion control.",
"best_for": "Controlled style with configured CustomVoice speakers.", "best_for": "Controlled style with configured CustomVoice speakers.",
"uses_wav": False, "style_aware": True, "true_streaming": False, "uses_wav": False, "style_aware": True, "true_streaming": False,
"speed": "~0.3× GPU", "latency": "1–3 s", "quality": "Premium", "ram": "6–8 GB VRAM", "speed": "~0.3× GPU", "latency": "1–3 s", "quality": "Premium", "ram": "~5 GB",
}, },
"streaming": { "streaming": {
"purpose": "Low-latency playback from saved WAV/reference voices.", "purpose": "Low-latency playback from saved WAV/reference voices.",
@ -3237,7 +3237,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Weak per-request style in the current streaming server.", "style": "Weak per-request style in the current streaming server.",
"best_for": "Long text, assistants, Open WebUI/SillyTavern playback that can start before completion.", "best_for": "Long text, assistants, Open WebUI/SillyTavern playback that can start before completion.",
"uses_wav": True, "style_aware": False, "true_streaming": True, "uses_wav": True, "style_aware": False, "true_streaming": True,
"speed": "~0.1× GPU", "latency": "0.5–1 s", "quality": "Premium", "ram": "6–8 GB VRAM", "speed": "~0.1× GPU", "latency": "0.5–1 s", "quality": "Premium", "ram": "~5 GB",
}, },
"nvidia_magpie": { "nvidia_magpie": {
"purpose": "Generate speech with NVIDIA Magpie fixed speaker voices.", "purpose": "Generate speech with NVIDIA Magpie fixed speaker voices.",
@ -3245,7 +3245,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Language and speaker are controlled by the backend voice config; per-request style text is usually ignored.", "style": "Language and speaker are controlled by the backend voice config; per-request style text is usually ignored.",
"best_for": "Fast local NVIDIA TTS voices and OpenAI-compatible assistant playback.", "best_for": "Fast local NVIDIA TTS voices and OpenAI-compatible assistant playback.",
"uses_wav": False, "style_aware": False, "true_streaming": False, "uses_wav": False, "style_aware": False, "true_streaming": False,
"speed": "~0.05× GPU", "latency": "0.3–0.8 s", "quality": "High", "ram": "4–6 GB VRAM", "speed": "~0.05× GPU", "latency": "0.3–0.8 s", "quality": "High", "ram": "~1 GB",
}, },
"nvidia_zeroshot": { "nvidia_zeroshot": {
"purpose": "Clone a saved library voice through NVIDIA Magpie TTS Zeroshot NIM.", "purpose": "Clone a saved library voice through NVIDIA Magpie TTS Zeroshot NIM.",
@ -3253,7 +3253,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Best with a clear 3-10 second prompt. Optional quality params can be configured in Settings.", "style": "Best with a clear 3-10 second prompt. Optional quality params can be configured in Settings.",
"best_for": "Fast NVIDIA reference-audio cloning, streaming-class use cases, live agents, and games.", "best_for": "Fast NVIDIA reference-audio cloning, streaming-class use cases, live agents, and games.",
"uses_wav": True, "style_aware": False, "true_streaming": False, "uses_wav": True, "style_aware": False, "true_streaming": False,
"speed": "~0.1× GPU", "latency": "0.5–1 s", "quality": "High clone", "ram": "4–6 GB VRAM", "speed": "~0.1× GPU", "latency": "0.5–1 s", "quality": "High clone", "ram": "~1 GB",
}, },
"nvidia_flow": { "nvidia_flow": {
"purpose": "Clone a saved library voice through NVIDIA Magpie TTS Flow NIM.", "purpose": "Clone a saved library voice through NVIDIA Magpie TTS Flow NIM.",
@ -3261,7 +3261,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Offline high-fidelity clone path; prompt transcript must match the reference audio.", "style": "Offline high-fidelity clone path; prompt transcript must match the reference audio.",
"best_for": "Studio-style dubbing, narration, and podcast-quality offline generation.", "best_for": "Studio-style dubbing, narration, and podcast-quality offline generation.",
"uses_wav": True, "style_aware": False, "true_streaming": False, "uses_wav": True, "style_aware": False, "true_streaming": False,
"speed": "~0.2× GPU", "latency": "1–2 s", "quality": "Studio", "ram": "4–6 GB VRAM", "speed": "~0.2× GPU", "latency": "1–2 s", "quality": "Studio", "ram": "~1 GB",
}, },
"kokoro": { "kokoro": {
"purpose": "High-quality English TTS with Kokoro 82M model. OpenAI-compatible endpoint.", "purpose": "High-quality English TTS with Kokoro 82M model. OpenAI-compatible endpoint.",
@ -3285,7 +3285,7 @@ def _backend_capabilities(backend: str) -> dict:
"style": "Speaker selected by voice ID. Style instruction not supported.", "style": "Speaker selected by voice ID. Style instruction not supported.",
"best_for": "Local multi-speaker TTS with XTTS v2 model. Coqui/daswer123 docker setup.", "best_for": "Local multi-speaker TTS with XTTS v2 model. Coqui/daswer123 docker setup.",
"uses_wav": False, "style_aware": False, "true_streaming": False, "uses_wav": False, "style_aware": False, "true_streaming": False,
"speed": "~0.3× GPU", "latency": "1–3 s", "quality": "High", "ram": "4–6 GB VRAM", "speed": "~0.3× GPU", "latency": "1–3 s", "quality": "High", "ram": "3–4 GB",
}, },
} }
return caps.get(_clean_preview_backend(backend), {}) return caps.get(_clean_preview_backend(backend), {})