Add unified prompt model support for multilingual ASR and streaming inference

examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streaming_infer.py

@@ -186,6 +186,16 @@ class TranscriptionConfig:
     # use_cer: bool = False
     debug_mode: bool = False  # Whether to print more detail in the output.
 
+    # Language-ID prompt for prompt-conditioned models (e.g. EncDecRNNTBPEModelWithPrompt).
+    # Set to a language key from the model's prompt_dictionary (e.g. "en-US", "auto").
+    # Ignored for models without prompt support.
+    target_lang: Optional[str] = None
+    # whether to strip the language tags from the transcriptions
+    # Ignored for model without prompt support
+    strip_lang_tags: bool = False
+    # Optional regex describing the language tag to strip. Defaults to "<xx-XX>". (r'\s*<[a-z]{2}-[A-Z]{2}>')
+    lang_tag_pattern: Optional[str] = None
+
 
 def extract_transcriptions(hyps):
     """
@@ -363,6 +373,12 @@ def main(cfg: TranscriptionConfig):
         else:
             asr_model.change_decoding_strategy(cfg.ctc_decoding)
 
+    # Set language-ID prompt for prompt-conditioned models
+    if hasattr(asr_model, 'set_inference_prompt'):
+        lang = cfg.target_lang if cfg.target_lang is not None else "auto"
+        asr_model.set_inference_prompt(lang)
+        asr_model.decoding.set_strip_lang_tags(cfg.strip_lang_tags, lang_tag_pattern=cfg.lang_tag_pattern)
+
     asr_model = asr_model.to(device=device, dtype=compute_dtype)
     asr_model.eval()
 

examples/asr/asr_transducer/speech_to_text_rnnt_bpe_prompt.py

@@ -0,0 +1,95 @@
+# Copyright (c) 2025, NVIDIA CORPORATION.  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.
+
+"""
+# Preparing the Tokenizer for the dataset
+Use the `process_asr_text_tokenizer.py` script under <NEMO_ROOT>/scripts/tokenizers/ in order to prepare the tokenizer.
+
+# Manifest file example:
+{"audio_filepath":"/data/audio.wav","duration":12.12,"text":"The transcript.","target_lang":"en-US"}
+
+```sh
+python <NEMO_ROOT>/scripts/tokenizers/process_asr_text_tokenizer.py \
+        --manifest=<path to train manifest files, seperated by commas>
+        OR
+        --data_file=<path to text data, seperated by commas> \
+        --data_root="<output directory>" \
+        --vocab_size=<number of tokens in vocabulary> \
+        --tokenizer=<"spe" or "wpe"> \
+        --no_lower_case \
+        --spe_type=<"unigram", "bpe", "char" or "word"> \
+        --spe_character_coverage=1.0 \
+        --log
+```
+
+# Training the model
+```sh
+python speech_to_text_rnnt_bpe_prompt.py \
+    # (Optional: --config-path=<path to dir of configs> --config-name=<name of config without .yaml>) \
+    model.train_ds.manifest_filepath=<path to train manifest> \
+    model.validation_ds.manifest_filepath=<path to val/test manifest> \
+    model.tokenizer.dir=<path to directory of tokenizer (not full path to the vocab file!)> \
+    model.tokenizer.type=<either bpe or wpe> \
+    trainer.devices=-1 \
+    trainer.max_epochs=100 \
+    model.optim.name="adamw" \
+    model.optim.lr=0.001 \
+    model.optim.betas=[0.9,0.999] \
+    model.optim.weight_decay=0.0001 \
+    model.optim.sched.warmup_steps=2000
+    exp_manager.create_wandb_logger=True \
+    exp_manager.wandb_logger_kwargs.name="<Name of experiment>" \
+    exp_manager.wandb_logger_kwargs.project="<Name of project>"
+```
+
+# Fine-tune a model
+
+For documentation on fine-tuning this model, please visit -
+https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/configs.html#fine-tuning-configurations
+
+"""
+
+import lightning.pytorch as pl
+from omegaconf import OmegaConf
+
+from nemo.collections.asr.models import EncDecRNNTBPEModelWithPrompt
+from nemo.core.config import hydra_runner
+from nemo.utils import logging
+from nemo.utils.exp_manager import exp_manager
+from nemo.utils.trainer_utils import resolve_trainer_cfg
+
+
+@hydra_runner(
+    config_path="../conf/fastconformer/cache_aware_streaming/",
+    config_name="fastconformer_transducer_bpe_streaming_prompt.yaml",
+)
+def main(cfg):
+    logging.info(f'Hydra config: {OmegaConf.to_yaml(cfg)}')
+
+    trainer = pl.Trainer(**resolve_trainer_cfg(cfg.trainer))
+    exp_manager(trainer, cfg.get("exp_manager", None))
+    asr_model = EncDecRNNTBPEModelWithPrompt(cfg=cfg.model, trainer=trainer)
+
+    # Initialize the weights of the model from another model, if provided via config
+    asr_model.maybe_init_from_pretrained_checkpoint(cfg)
+
+    trainer.fit(asr_model)
+
+    if hasattr(cfg.model, 'test_ds') and cfg.model.test_ds.manifest_filepath is not None:
+        if asr_model.prepare_test(trainer):
+            trainer.test(asr_model)
+
+
+if __name__ == '__main__':
+    main()  # noqa pylint: disable=no-value-for-parameter