Add C++ runtime and Python API for NeMo Canary models (#2352)

sherpa-onnx/python/sherpa_onnx/offline_recognizer.py

@@ -6,6 +6,7 @@ from typing import List, Optional
 from _sherpa_onnx import (
     FeatureExtractorConfig,
     HomophoneReplacerConfig,
+    OfflineCanaryModelConfig,
     OfflineCtcFstDecoderConfig,
     OfflineDolphinModelConfig,
     OfflineFireRedAsrModelConfig,
@@ -425,7 +426,6 @@ class OfflineRecognizer(object):
             num_threads=num_threads,
             debug=debug,
             provider=provider,
-            model_type="nemo_ctc",
         )
 
         feat_config = FeatureExtractorConfig(
@@ -690,6 +690,102 @@ class OfflineRecognizer(object):
         self.config = recognizer_config
         return self
 
+    @classmethod
+    def from_nemo_canary(
+        cls,
+        encoder: str,
+        decoder: str,
+        tokens: str,
+        src_lang: str = "en",
+        tgt_lang: str = "en",
+        num_threads: int = 1,
+        sample_rate: int = 16000,
+        feature_dim: int = 128,  # not used
+        decoding_method: str = "greedy_search",  # not used
+        debug: bool = False,
+        provider: str = "cpu",
+        rule_fsts: str = "",
+        rule_fars: str = "",
+        hr_dict_dir: str = "",
+        hr_rule_fsts: str = "",
+        hr_lexicon: str = "",
+    ):
+        """
+        Please refer to
+        `<https://k2-fsa.github.io/sherpa/onnx/nemo/index.html>`_
+        to download pre-trained models for different languages.
+
+        Args:
+          encoder:
+            Path to ``encoder.onnx`` or ``encoder.int8.onnx``.
+          decoder:
+            Path to ``decoder.onnx`` or ``decoder.int8.onnx``.
+          tokens:
+            Path to ``tokens.txt``. Each line in ``tokens.txt`` contains two
+            columns::
+
+                symbol integer_id
+
+          src_lang:
+            The language of the input audio. Valid values are: en, es, de, fr.
+            If you leave it empty, it uses en internally.
+          tgt_lang:
+            The language of the output text. Valid values are: en, es, de, fr.
+            If you leave it empty, it uses en internally.
+          num_threads:
+            Number of threads for neural network computation.
+          sample_rate:
+            Sample rate of the training data used to train the model. Not used
+          feature_dim:
+            Dimension of the feature used to train the model. Not used
+          decoding_method:
+            Valid values are greedy_search. Not used
+          debug:
+            True to show debug messages.
+          provider:
+            onnxruntime execution providers. Valid values are: cpu, cuda, coreml.
+          rule_fsts:
+            If not empty, it specifies fsts for inverse text normalization.
+            If there are multiple fsts, they are separated by a comma.
+          rule_fars:
+            If not empty, it specifies fst archives for inverse text normalization.
+            If there are multiple archives, they are separated by a comma.
+        """
+        self = cls.__new__(cls)
+        model_config = OfflineModelConfig(
+            canary=OfflineCanaryModelConfig(
+                encoder=encoder,
+                decoder=decoder,
+                src_lang=src_lang,
+                tgt_lang=tgt_lang,
+            ),
+            tokens=tokens,
+            num_threads=num_threads,
+            debug=debug,
+            provider=provider,
+        )
+
+        feat_config = FeatureExtractorConfig(
+            sampling_rate=sample_rate,
+            feature_dim=feature_dim,
+        )
+
+        recognizer_config = OfflineRecognizerConfig(
+            feat_config=feat_config,
+            model_config=model_config,
+            decoding_method=decoding_method,
+            rule_fsts=rule_fsts,
+            rule_fars=rule_fars,
+            hr=HomophoneReplacerConfig(
+                dict_dir=hr_dict_dir,
+                lexicon=hr_lexicon,
+                rule_fsts=hr_rule_fsts,
+            ),
+        )
+        self.recognizer = _Recognizer(recognizer_config)
+        self.config = recognizer_config
+        return self
+
     @classmethod
     def from_whisper(
         cls,