Add MOSS-TTS-v1.5 (delay) and MOSS-Audio-Tokenizer

docs/source/en/_toctree.yml

@@ -1105,6 +1105,10 @@
         title: Moonshine Streaming
       - local: model_doc/moshi
         title: Moshi
+      - local: model_doc/moss-audio-tokenizer
+        title: MOSS Audio Tokenizer
+      - local: model_doc/moss_tts_delay
+        title: MOSS-TTS Delay
       - local: model_doc/musicgen
         title: MusicGen
       - local: model_doc/musicgen_melody

docs/source/en/model_doc/moss-audio-tokenizer.md

@@ -0,0 +1,66 @@
+<!--Copyright 2026 OpenMOSS and The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+*This model was contributed to Hugging Face Transformers on 2026-06-14.*
+
+# MOSS Audio Tokenizer
+
+[MOSS-Audio-Tokenizer](https://huggingface.co/OpenMOSS-Team/MOSS-Audio-Tokenizer) is the neural audio codec used by
+MOSS-TTS. It encodes waveforms into discrete audio codebook tokens and decodes those tokens back into waveform audio.
+
+## Usage
+
+```python
+import torch
+
+from transformers import MossAudioTokenizerModel
+
+
+model_id = "OpenMOSS-Team/MOSS-Audio-Tokenizer"
+model = MossAudioTokenizerModel.from_pretrained(model_id, device_map="auto")
+
+audio = torch.randn(1, 1, 24000, device=model.device)
+
+encoded = model.encode(audio, return_dict=True)
+audio_codes = encoded.audio_codes
+
+decoded = model.decode(audio_codes, return_dict=True)
+audio_values = decoded.audio
+```
+
+## MossAudioTokenizerConfig
+
+[[autodoc]] MossAudioTokenizerConfig
+
+## MossAudioTokenizerEncoderConfig
+
+[[autodoc]] MossAudioTokenizerEncoderConfig
+
+## MossAudioTokenizerDecoderConfig
+
+[[autodoc]] MossAudioTokenizerDecoderConfig
+
+## MossAudioTokenizerQuantizerConfig
+
+[[autodoc]] MossAudioTokenizerQuantizerConfig
+
+## MossAudioTokenizerModel
+
+[[autodoc]] MossAudioTokenizerModel
+    - encode
+    - decode
+    - batch_encode
+    - batch_decode
+    - forward

docs/source/en/model_doc/moss_tts_delay.md

@@ -0,0 +1,72 @@
+<!--Copyright 2026 OpenMOSS and The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+*This model was contributed to Hugging Face Transformers on 2026-06-14.*
+
+# MOSS-TTS Delay
+
+[MOSS-TTS-v1.5](https://huggingface.co/OpenMOSS-Team/MOSS-TTS-v1.5) is a multilingual text-to-speech model from
+OpenMOSS. The model uses a Qwen3 language backbone and predicts delayed audio codebook tokens for speech generation.
+
+## Usage
+
+```python
+from scipy.io.wavfile import write
+
+from transformers import AutoProcessor, AutoModelForTextToWaveform
+
+
+model_id = "OpenMOSS-Team/MOSS-TTS-v1.5"
+
+processor = AutoProcessor.from_pretrained(model_id)
+model = AutoModelForTextToWaveform.from_pretrained(model_id, dtype="auto", device_map="auto")
+
+message = processor.build_user_message(text="Hello from MOSS-TTS.", language="English")
+inputs = processor([message]).to(model.device)
+outputs = model.generate(**inputs, max_new_tokens=1024)
+
+messages = processor.decode(outputs)
+audio_values = messages[0].audio_codes_list[0]
+write("moss_tts.wav", processor.model_config.sampling_rate, audio_values.cpu().numpy())
+```
+
+## MossTTSDelayConfig
+
+[[autodoc]] MossTTSDelayConfig
+
+## MossTTSDelayProcessor
+
+[[autodoc]] MossTTSDelayProcessor
+
+## Message
+
+[[autodoc]] Message
+
+## UserMessage
+
+[[autodoc]] UserMessage
+
+## AssistantMessage
+
+[[autodoc]] AssistantMessage
+
+## MossTTSDelayModel
+
+[[autodoc]] MossTTSDelayModel
+    - forward
+
+## MossTTSDelayOutputWithPast
+
+[[autodoc]] MossTTSDelayOutputWithPast

src/transformers/models/__init__.py

@@ -286,6 +286,8 @@
     from .moonshine import *
     from .moonshine_streaming import *
     from .moshi import *
+    from .moss_audio_tokenizer import *
+    from .moss_tts_delay import *
     from .mpnet import *
     from .mpt import *
     from .mra import *

src/transformers/models/auto/auto_mappings.py

@@ -382,6 +382,11 @@
         ("moonshine_streaming_encoder", "MoonshineStreamingEncoderConfig"),
         ("moshi", "MoshiConfig"),
         ("moshi_depth", "MoshiDepthConfig"),
+        ("moss-audio-tokenizer", "MossAudioTokenizerConfig"),
+        ("moss-audio-tokenizer-decoder", "MossAudioTokenizerDecoderConfig"),
+        ("moss-audio-tokenizer-encoder", "MossAudioTokenizerEncoderConfig"),
+        ("moss-audio-tokenizer-quantizer", "MossAudioTokenizerQuantizerConfig"),
+        ("moss_tts_delay", "MossTTSDelayConfig"),
         ("mpnet", "MPNetConfig"),
         ("mpt", "MptConfig"),
         ("mra", "MraConfig"),
@@ -793,6 +798,10 @@
         ("modernbert-decoder", "modernbert_decoder"),
         ("moonshine_streaming_encoder", "moonshine_streaming"),
         ("moshi_depth", "moshi"),
+        ("moss-audio-tokenizer", "moss_audio_tokenizer"),
+        ("moss-audio-tokenizer-decoder", "moss_audio_tokenizer"),
+        ("moss-audio-tokenizer-encoder", "moss_audio_tokenizer"),
+        ("moss-audio-tokenizer-quantizer", "moss_audio_tokenizer"),
         ("musicgen_decoder", "musicgen"),
         ("musicgen_melody_decoder", "musicgen_melody"),
         ("nllb-moe", "nllb_moe"),
@@ -1022,6 +1031,7 @@
         ("minicpmv4_6", "MiniCPMV4_6Processor"),
         ("mllama", "MllamaProcessor"),
         ("moonshine_streaming", "MoonshineStreamingProcessor"),
+        ("moss_tts_delay", "MossTTSDelayProcessor"),
         ("musicflamingo", "MusicFlamingoProcessor"),
         ("musicgen", "MusicgenProcessor"),
         ("musicgen_melody", "MusicgenMelodyProcessor"),

src/transformers/models/auto/modeling_auto.py

@@ -322,6 +322,8 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         ("moonshine", "MoonshineModel"),
         ("moonshine_streaming", "MoonshineStreamingModel"),
         ("moshi", "MoshiModel"),
+        ("moss-audio-tokenizer", "MossAudioTokenizerModel"),
+        ("moss_tts_delay", "MossTTSDelayModel"),
         ("mpnet", "MPNetModel"),
         ("mpt", "MptModel"),
         ("mra", "MraModel"),
@@ -1746,6 +1748,7 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         ("csm", "CsmForConditionalGeneration"),
         ("fastspeech2_conformer_with_hifigan", "FastSpeech2ConformerWithHifiGan"),
         ("higgs_audio_v2", "HiggsAudioV2ForConditionalGeneration"),
+        ("moss_tts_delay", "MossTTSDelayModel"),
         ("musicgen", "MusicgenForConditionalGeneration"),
         ("musicgen_melody", "MusicgenMelodyForConditionalGeneration"),
         ("qwen2_5_omni", "Qwen2_5OmniForConditionalGeneration"),
@@ -1903,6 +1906,7 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
     [
         ("dac", "DacModel"),
         ("higgs_audio_v2_tokenizer", "HiggsAudioV2TokenizerModel"),
+        ("moss-audio-tokenizer", "MossAudioTokenizerModel"),
         ("vibevoice_acoustic_tokenizer", "VibeVoiceAcousticTokenizerModel"),
     ]
 )

src/transformers/models/auto/tokenization_auto.py

@@ -220,6 +220,7 @@
         ("mluke", "MLukeTokenizer" if is_sentencepiece_available() else None),
         ("mm-grounding-dino", "BertTokenizer" if is_tokenizers_available() else None),
         ("mobilebert", "MobileBertTokenizer" if is_tokenizers_available() else None),
+        ("moss_tts_delay", "Qwen2Tokenizer" if is_tokenizers_available() else None),
         ("mpnet", "MPNetTokenizer" if is_tokenizers_available() else None),
         ("mpt", "GPTNeoXTokenizer" if is_tokenizers_available() else None),
         ("mra", "RobertaTokenizer"),

src/transformers/models/moss_audio_tokenizer/__init__.py

@@ -0,0 +1,27 @@
+# Copyright 2026 OpenMOSS and the HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import TYPE_CHECKING
+
+from ...utils import _LazyModule
+from ...utils.import_utils import define_import_structure
+
+
+if TYPE_CHECKING:
+    from .configuration_moss_audio_tokenizer import *
+    from .modeling_moss_audio_tokenizer import *
+else:
+    import sys
+
+    _file = globals()["__file__"]
+    sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)