feat(chatterbox): support Turbo and Nano

.github/workflows/lint.yml

@@ -130,3 +130,5 @@ jobs:
       # ride along here instead of waiting on a multi-GB backend image build.
       - name: run shared python backend helper tests
         run: make test-python-helpers
+      - name: run Chatterbox variant tests
+        run: make -C backend/python/chatterbox test-unit

backend/python/chatterbox/Makefile

@@ -14,10 +14,14 @@ test: chatterbox
 	bash test.sh
 	@echo "coqui tested."
 
+.PHONY: test-unit
+test-unit:
+	python3 -m unittest discover -s . -p 'test_variants.py' -v
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py
 
 .PHONY: clean
 clean: protogen-clean
-	rm -rf venv __pycache__
+	rm -rf venv __pycache__

backend/python/chatterbox/backend.py

@@ -38,16 +38,18 @@ def is_int(s):
     except ValueError:
         return False
 
-def coerce_param_value(value):
+def coerce_param_value(value, key=None):
     """Coerce a TTSRequest.params value (string on the wire) to the type the
     Chatterbox generate() kwargs expect (float/int/bool), matching how static
     YAML options are coerced at load time. Non-string values pass through."""
     if not isinstance(value, str):
         return value
+    # Transformers requires top_k to be an int, but float controls such as
+    # temperature must remain floats even when written without a decimal point.
+    if key == "top_k" and is_int(value):
+        return int(value)
     if is_float(value):
         return float(value)
-    if is_int(value):
-        return int(value)
     if value.lower() in ["true", "false"]:
         return value.lower() == "true"
     return value
@@ -157,14 +159,7 @@ def LoadModel(self, request, context):
             if ":" not in opt:
                 continue
             key, value = opt.split(":")
-            # if value is a number, convert it to the appropriate type
-            if is_float(value):
-                value = float(value)
-            elif is_int(value):
-                value = int(value)
-            elif value.lower() in ["true", "false"]:
-                value = value.lower() == "true"
-            self.options[key] = value
+            self.options[key] = coerce_param_value(value, key)
 
         self.AudioPath = None
 
@@ -177,9 +172,23 @@ def LoadModel(self, request, context):
             self.AudioPath = os.path.join(modelFileBase, request.AudioPath)
         try:
             print("Preparing models, please wait", file=sys.stderr)
-            if "multilingual" in self.options:
-                # remove key from options
-                del self.options["multilingual"]
+            selectors = {}
+            for name in ("multilingual", "turbo", "nano"):
+                value = self.options.pop(name, False)
+                if not isinstance(value, bool):
+                    raise ValueError(f"{name} must be true or false")
+                selectors[name] = value
+            if sum(selectors.values()) > 1:
+                raise ValueError("Only one of multilingual, turbo, or nano can be true")
+            self.variant = next((name for name, enabled in selectors.items() if enabled), "original")
+            if self.variant in ("turbo", "nano"):
+                # Older installations can still load the original models.
+                from chatterbox.tts_turbo import ChatterboxTurboTTS
+                if self.variant == "nano":
+                    self.model = ChatterboxTurboTTS.from_pretrained(device=device, nano=True)
+                else:
+                    self.model = ChatterboxTurboTTS.from_pretrained(device=device)
+            elif self.variant == "multilingual":
                 self.model = ChatterboxMultilingualTTS.from_pretrained(device=device)
             else:
                 self.model = ChatterboxTTS.from_pretrained(device=device)
@@ -211,7 +220,15 @@ def TTS(self, request, context):
                 for key, value in request.params.items():
                     if key == "ref_text":
                         continue
-                    kwargs[key] = coerce_param_value(value)
+                    kwargs[key] = coerce_param_value(value, key)
+
+            # Selectors apply only at load time, including when supplied in params.
+            for name in ("multilingual", "turbo", "nano"):
+                kwargs.pop(name, None)
+            if self.variant in ("turbo", "nano"):
+                # These controls belong to the original model's sampler.
+                for name in ("cfg_weight", "exaggeration", "min_p"):
+                    kwargs.pop(name, None)
 
             # Check if text exceeds 250 characters
             # (chatterbox does not support long text)

backend/python/chatterbox/install.sh

@@ -29,9 +29,11 @@ installRequirements
 # pip to backtrack through Jinja2/MarkupSafe/omegaconf/ruamel.yaml into Python-2-era
 # sdists that no longer build. We install chatterbox-tts itself with --no-deps and
 # list its real runtime deps in requirements-*.txt instead.
+# Pin the fork so every backend build includes the same Turbo/Nano implementation.
+CHATTERBOX_REV=b38c317b03834ef35828ae96c72ac8929b002a05
 echo "Installing chatterbox-tts with --no-deps"
 if [ "x${USE_PIP}" == "xtrue" ]; then
-    pip install ${EXTRA_PIP_INSTALL_FLAGS:-} --no-deps "chatterbox-tts@git+https://git@github.com/mudler/chatterbox.git@faster"
+    pip install ${EXTRA_PIP_INSTALL_FLAGS:-} --no-deps "chatterbox-tts@git+https://git@github.com/mudler/chatterbox.git@${CHATTERBOX_REV}"
 else
-    uv pip install ${EXTRA_PIP_INSTALL_FLAGS:-} --no-deps "chatterbox-tts@git+https://git@github.com/mudler/chatterbox.git@faster"
+    uv pip install ${EXTRA_PIP_INSTALL_FLAGS:-} --no-deps "chatterbox-tts@git+https://git@github.com/mudler/chatterbox.git@${CHATTERBOX_REV}"
 fi

backend/python/chatterbox/requirements-cpu.txt

@@ -3,7 +3,7 @@ accelerate
 torch
 torchaudio
 numpy>=1.24.0,<1.26.0
-transformers
+transformers==4.46.3
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml
 # minus russian-text-stresser (whose ancient pins break the resolver).
@@ -16,4 +16,4 @@ resemble-perth==1.0.1
 conformer
 safetensors
 spacy-pkuseg
-pykakasi==2.3.0
+pykakasi==2.3.0

backend/python/chatterbox/requirements-cublas12.txt

@@ -1,7 +1,7 @@
 --extra-index-url https://download.pytorch.org/whl/cu124
 torch==2.6.0+cu124
 torchaudio==2.6.0+cu124
-transformers<5
+transformers==4.46.3
 numpy>=1.24.0,<1.26.0
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml
@@ -16,4 +16,4 @@ conformer
 safetensors
 spacy-pkuseg
 pykakasi==2.3.0
-accelerate
+accelerate

backend/python/chatterbox/requirements-cublas13.txt

@@ -1,7 +1,7 @@
 --extra-index-url https://download.pytorch.org/whl/cu130
 torch
 torchaudio
-transformers
+transformers==4.46.3
 numpy>=1.24.0,<1.26.0
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml

backend/python/chatterbox/requirements-hipblas.txt

@@ -1,7 +1,7 @@
 --extra-index-url https://download.pytorch.org/whl/rocm7.0
 torch==2.10.0+rocm7.0
 torchaudio==2.10.0+rocm7.0
-transformers
+transformers==4.46.3
 numpy>=1.24.0,<1.26.0
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml

backend/python/chatterbox/requirements-intel.txt

@@ -1,7 +1,7 @@
 --extra-index-url https://download.pytorch.org/whl/xpu
 torch
 torchaudio
-transformers
+transformers==4.46.3
 numpy>=1.24.0,<1.26.0
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml
@@ -19,4 +19,4 @@ pykakasi==2.3.0
 accelerate
 oneccl_bind_pt==2.3.100+xpu
 optimum[openvino]
-setuptools
+setuptools

backend/python/chatterbox/requirements-l4t12.txt

@@ -1,7 +1,7 @@
 --extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu126/
 torch
 torchaudio
-transformers
+transformers==4.46.3
 numpy>=1.24.0,<1.26.0
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml

backend/python/chatterbox/requirements-l4t13.txt

@@ -1,7 +1,7 @@
 --extra-index-url https://download.pytorch.org/whl/cu130
 torch
 torchaudio
-transformers
+transformers==4.46.3
 numpy>=1.24.0,<1.26.0
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml

backend/python/chatterbox/requirements-mps.txt

@@ -2,7 +2,7 @@ torch
 torchaudio
 accelerate
 numpy>=1.24.0,<1.26.0
-transformers
+transformers==4.46.3
 # chatterbox-tts itself is installed with --no-deps in install.sh.
 # These are its real runtime deps, mirroring upstream's pyproject.toml
 # minus russian-text-stresser (whose ancient pins break the resolver).
@@ -15,4 +15,4 @@ resemble-perth==1.0.1
 conformer
 safetensors
 spacy-pkuseg
-pykakasi==2.3.0
+pykakasi==2.3.0

backend/python/chatterbox/requirements.txt

@@ -3,4 +3,6 @@ protobuf
 certifi
 packaging
 setuptools<81
-poetry
+poetry
+# Turbo and Nano normalize reference audio before cloning.
+pyloudnorm