4413868535

Build Type

push

github

Committed by GitHub

Commit Message

SPARKNLP-746: Handle empty validation sets (#13615)

Run Details

8597 of 12936 relevant lines covered (66.46%)

0.66 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

0.0

/src/main/scala/com/johnsnowlabs/nlp/annotators/audio/HubertForCTC.scala

/*
 * Copyright 2017-2022 John Snow Labs
 *
 * 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.
 */

package com.johnsnowlabs.nlp.annotators.audio

import com.johnsnowlabs.ml.tensorflow.{ReadTensorflowModel, TensorflowWrapper}
import com.johnsnowlabs.ml.util.LoadExternalModel.{
  loadJsonStringAsset,
  modelSanityCheck,
  notSupportedEngineError
}
import com.johnsnowlabs.ml.util.ModelEngine
import com.johnsnowlabs.nlp._
import com.johnsnowlabs.nlp.annotators.audio.feature_extractor.Preprocessor
import org.apache.spark.ml.util.Identifiable
import org.apache.spark.sql.SparkSession
import org.json4s._
import org.json4s.jackson.JsonMethods._

/** Hubert Model with a language modeling head on top for Connectionist Temporal Classification
  * (CTC). Hubert was proposed in HuBERT: Self-Supervised Speech Representation Learning by Masked
  * Prediction of Hidden Units by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal
  * Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
  *
  * The annotator takes audio files and transcribes it as text. The audio needs to be provided
  * pre-processed an array of floats.
  *
  * Note that this annotator is currently not supported on Apple Silicon processors such as the
  * M1/M2 (Apple Silicon). This is due to the processor not supporting instructions for XLA.
  *
  * Pretrained models can be loaded with `pretrained` of the companion object:
  * {{{
  * val speechToText = HubertForCTC.pretrained()
  *   .setInputCols("audio_assembler")
  *   .setOutputCol("text")
  * }}}
  * The default model is `"asr_hubert_large_ls960"`, if no name is provided.
  *
  * For available pretrained models please see the
  * [[https://nlp.johnsnowlabs.com/models Models Hub]].
  *
  * To see which models are compatible and how to import them see
  * [[https://github.com/JohnSnowLabs/spark-nlp/discussions/5669]] and to see more extended
  * examples, see
  * [[https://github.com/JohnSnowLabs/spark-nlp/blob/master/src/test/scala/com/johnsnowlabs/nlp/annotators/audio/HubertForCTCTestSpec.scala HubertForCTCTestSpec]].
  *
  * '''References:'''
  *
  * [[https://arxiv.org/abs/2106.07447 HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units]]
  *
  * '''Paper Abstract:'''
  *
  * ''Self-supervised approaches for speech representation learning are challenged by three unique
  * problems: (1) there are multiple sound units in each input utterance, (2) there is no lexicon
  * of input sound units during the pre-training phase, and (3) sound units have variable lengths
  * with no explicit segmentation. To deal with these three problems, we propose the Hidden-Unit
  * BERT (HuBERT) approach for self-supervised speech representation learning, which utilizes an
  * offline clustering step to provide aligned target labels for a BERT-like prediction loss. A
  * key ingredient of our approach is applying the prediction loss over the masked regions only,
  * which forces the model to learn a combined acoustic and language model over the continuous
  * inputs. HuBERT relies primarily on the consistency of the unsupervised clustering step rather
  * than the intrinsic quality of the assigned cluster labels. Starting with a simple k-means
  * teacher of 100 clusters, and using two iterations of clustering, the HuBERT model either
  * matches or improves upon the state-of-the-art wav2vec 2.0 performance on the Librispeech
  * (960h) and Libri-light (60,000h) benchmarks with 10min, 1h, 10h, 100h, and 960h fine-tuning
  * subsets. Using a 1B parameter model, HuBERT shows up to 19% and 13% relative WER reduction on
  * the more challenging dev-other and test-other evaluation subsets.''
  *
  * ==Example==
  * {{{
  * import spark.implicits._
  * import com.johnsnowlabs.nlp.base._
  * import com.johnsnowlabs.nlp.annotators._
  * import com.johnsnowlabs.nlp.annotators.audio.HubertForCTC
  * import org.apache.spark.ml.Pipeline
  *
  * val audioAssembler: AudioAssembler = new AudioAssembler()
  *   .setInputCol("audio_content")
  *   .setOutputCol("audio_assembler")
  *
  * val speechToText: HubertForCTC = HubertForCTC
  *   .pretrained()
  *   .setInputCols("audio_assembler")
  *   .setOutputCol("text")
  *
  * val pipeline: Pipeline = new Pipeline().setStages(Array(audioAssembler, speechToText))
  *
  * val bufferedSource =
  *   scala.io.Source.fromFile("src/test/resources/audio/csv/audio_floats.csv")
  *
  * val rawFloats = bufferedSource
  *   .getLines()
  *   .map(_.split(",").head.trim.toFloat)
  *   .toArray
  * bufferedSource.close
  *
  * val processedAudioFloats = Seq(rawFloats).toDF("audio_content")
  *
  * val result = pipeline.fit(processedAudioFloats).transform(processedAudioFloats)
  * result.select("text.result").show(truncate = false)
  * +------------------------------------------------------------------------------------------+
  * |result                                                                                    |
  * +------------------------------------------------------------------------------------------+
  * |[MISTER QUILTER IS THE APOSTLE OF THE MIDLE CLASES AND WE ARE GLAD TO WELCOME HIS GOSPEL ]|
  * +------------------------------------------------------------------------------------------+
  * }}}
  *
  * @param uid
  *   required uid for storing annotator to disk
  * @groupname anno Annotator types
  * @groupdesc anno
  *   Required input and expected output annotator types
  * @groupname Ungrouped Members
  * @groupname param Parameters
  * @groupname setParam Parameter setters
  * @groupname getParam Parameter getters
  * @groupname Ungrouped Members
  * @groupprio param  1
  * @groupprio anno  2
  * @groupprio Ungrouped 3
  * @groupprio setParam  4
  * @groupprio getParam  5
  * @groupdesc param
  *   A list of (hyper-)parameter keys this annotator can take. Users can set and get the
  *   parameter values through setters and getters, respectively.
  */
class HubertForCTC(override val uid: String) extends Wav2Vec2ForCTC(uid) {

  /** Annotator reference id. Used to identify elements in metadata or to refer to this annotator
    * type
    */
  def this() = this(Identifiable.randomUID("HubertForCTC"))

  override def onWrite(path: String, spark: SparkSession): Unit = {
    super.onWrite(path, spark)
    writeTensorflowModelV2(
      path,
      spark,
      getModelIfNotSet.tensorflowWrapper,
      "_hubert_ctc",
      HubertForCTC.tfFile,
      configProtoBytes = getConfigProtoBytes)
  }

}

trait ReadablePretrainedHubertForAudioModel
    extends ParamsAndFeaturesReadable[HubertForCTC]
    with HasPretrained[HubertForCTC] {
  override val defaultModelName: Some[String] = Some("asr_hubert_large_ls960")

  /** Java compliant-overrides */
  override def pretrained(): HubertForCTC = super.pretrained()

  override def pretrained(name: String): HubertForCTC = super.pretrained(name)

  override def pretrained(name: String, lang: String): HubertForCTC =
    super.pretrained(name, lang)

  override def pretrained(name: String, lang: String, remoteLoc: String): HubertForCTC =
    super.pretrained(name, lang, remoteLoc)
}

trait ReadHubertForAudioDLModel extends ReadTensorflowModel {
  this: ParamsAndFeaturesReadable[HubertForCTC] =>

  override val tfFile: String = "hubert_ctc_tensorflow"

  def readTensorflow(instance: HubertForCTC, path: String, spark: SparkSession): Unit = {

    val tf = readTensorflowModel(path, spark, "_hubert_ctc_tf", initAllTables = false)
    instance.setModelIfNotSet(spark, tf)
  }

  addReader(readTensorflow)

  def loadSavedModel(modelPath: String, spark: SparkSession): HubertForCTC = {

    val (localModelPath, detectedEngine) = modelSanityCheck(modelPath)

    val vocabJsonContent = loadJsonStringAsset(localModelPath, "vocab.json")
    val vocabJsonMap =
      parse(vocabJsonContent, useBigIntForLong = true).values
        .asInstanceOf[Map[String, BigInt]]

    val preprocessorConfigJsonContent =
      loadJsonStringAsset(localModelPath, "preprocessor_config.json")
    val preprocessorConfig =
      Preprocessor.loadPreprocessorConfig(preprocessorConfigJsonContent)

    /*Universal parameters for all engines*/
    val annotatorModel = new HubertForCTC()
      .setVocabulary(vocabJsonMap)
      .setDoNormalize(preprocessorConfig.do_normalize)
      .setFeatureSize(preprocessorConfig.feature_size)
      .setPaddingSide(preprocessorConfig.padding_side)
      .setPaddingValue(preprocessorConfig.padding_value)
      .setReturnAttentionMask(preprocessorConfig.return_attention_mask)
      .setSamplingRate(preprocessorConfig.sampling_rate)

    annotatorModel.set(annotatorModel.engine, detectedEngine)

    detectedEngine match {
      case ModelEngine.tensorflow =>
        val (wrapper, signatures) =
          TensorflowWrapper.read(localModelPath, zipped = false, useBundle = true)

        val _signatures = signatures match {
          case Some(s) => s
          case None => throw new Exception("Cannot load signature definitions from model!")
        }

        /** the order of setSignatures is important if we use getSignatures inside
          * setModelIfNotSet
          */
        annotatorModel
          .setSignatures(_signatures)
          .setModelIfNotSet(spark, wrapper)

      case _ =>
        throw new Exception(notSupportedEngineError)
    }

    annotatorModel
  }
}

/** This is the companion object of [[HubertForCTC]]. Please refer to that class for the
  * documentation.
  */
object HubertForCTC extends ReadablePretrainedHubertForAudioModel with ReadHubertForAudioDLModel

1	/*
2	* Copyright 2017-2022 John Snow Labs
3	*
4	* Licensed under the Apache License, Version 2.0 (the "License");
5	* you may not use this file except in compliance with the License.
6	* You may obtain a copy of the License at
7	*
8	* http://www.apache.org/licenses/LICENSE-2.0
9	*
10	* Unless required by applicable law or agreed to in writing, software
11	* distributed under the License is distributed on an "AS IS" BASIS,
12	* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13	* See the License for the specific language governing permissions and
14	* limitations under the License.
15	*/
16
17	package com.johnsnowlabs.nlp.annotators.audio
18
19	import com.johnsnowlabs.ml.tensorflow.{ReadTensorflowModel, TensorflowWrapper}
20	import com.johnsnowlabs.ml.util.LoadExternalModel.{
21	loadJsonStringAsset,
22	modelSanityCheck,
23	notSupportedEngineError
24	}
25	import com.johnsnowlabs.ml.util.ModelEngine
26	import com.johnsnowlabs.nlp._
27	import com.johnsnowlabs.nlp.annotators.audio.feature_extractor.Preprocessor
28	import org.apache.spark.ml.util.Identifiable
29	import org.apache.spark.sql.SparkSession
30	import org.json4s._
31	import org.json4s.jackson.JsonMethods._
32
33	/** Hubert Model with a language modeling head on top for Connectionist Temporal Classification
34	* (CTC). Hubert was proposed in HuBERT: Self-Supervised Speech Representation Learning by Masked
35	* Prediction of Hidden Units by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal
36	* Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
37	*
38	* The annotator takes audio files and transcribes it as text. The audio needs to be provided
39	* pre-processed an array of floats.
40	*
41	* Note that this annotator is currently not supported on Apple Silicon processors such as the
42	* M1/M2 (Apple Silicon). This is due to the processor not supporting instructions for XLA.
43	*
44	* Pretrained models can be loaded with `pretrained` of the companion object:
45	* {{{
46	* val speechToText = HubertForCTC.pretrained()
47	* .setInputCols("audio_assembler")
48	* .setOutputCol("text")
49	* }}}
50	* The default model is `"asr_hubert_large_ls960"`, if no name is provided.
51	*
52	* For available pretrained models please see the
53	* [[https://nlp.johnsnowlabs.com/models Models Hub]].
54	*
55	* To see which models are compatible and how to import them see
56	* [[https://github.com/JohnSnowLabs/spark-nlp/discussions/5669]] and to see more extended
57	* examples, see
58	* [[https://github.com/JohnSnowLabs/spark-nlp/blob/master/src/test/scala/com/johnsnowlabs/nlp/annotators/audio/HubertForCTCTestSpec.scala HubertForCTCTestSpec]].
59	*
60	* '''References:'''
61	*
62	* [[https://arxiv.org/abs/2106.07447 HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units]]
63	*
64	* '''Paper Abstract:'''
65	*
66	* ''Self-supervised approaches for speech representation learning are challenged by three unique
67	* problems: (1) there are multiple sound units in each input utterance, (2) there is no lexicon
68	* of input sound units during the pre-training phase, and (3) sound units have variable lengths
69	* with no explicit segmentation. To deal with these three problems, we propose the Hidden-Unit
70	* BERT (HuBERT) approach for self-supervised speech representation learning, which utilizes an
71	* offline clustering step to provide aligned target labels for a BERT-like prediction loss. A
72	* key ingredient of our approach is applying the prediction loss over the masked regions only,
73	* which forces the model to learn a combined acoustic and language model over the continuous
74	* inputs. HuBERT relies primarily on the consistency of the unsupervised clustering step rather
75	* than the intrinsic quality of the assigned cluster labels. Starting with a simple k-means
76	* teacher of 100 clusters, and using two iterations of clustering, the HuBERT model either
77	* matches or improves upon the state-of-the-art wav2vec 2.0 performance on the Librispeech
78	* (960h) and Libri-light (60,000h) benchmarks with 10min, 1h, 10h, 100h, and 960h fine-tuning
79	* subsets. Using a 1B parameter model, HuBERT shows up to 19% and 13% relative WER reduction on
80	* the more challenging dev-other and test-other evaluation subsets.''
81	*
82	* ==Example==
83	* {{{
84	* import spark.implicits._
85	* import com.johnsnowlabs.nlp.base._
86	* import com.johnsnowlabs.nlp.annotators._
87	* import com.johnsnowlabs.nlp.annotators.audio.HubertForCTC
88	* import org.apache.spark.ml.Pipeline
89	*
90	* val audioAssembler: AudioAssembler = new AudioAssembler()
91	* .setInputCol("audio_content")
92	* .setOutputCol("audio_assembler")
93	*
94	* val speechToText: HubertForCTC = HubertForCTC
95	* .pretrained()
96	* .setInputCols("audio_assembler")
97	* .setOutputCol("text")
98	*
99	* val pipeline: Pipeline = new Pipeline().setStages(Array(audioAssembler, speechToText))
100	*
101	* val bufferedSource =
102	* scala.io.Source.fromFile("src/test/resources/audio/csv/audio_floats.csv")
103	*
104	* val rawFloats = bufferedSource
105	* .getLines()
106	* .map(_.split(",").head.trim.toFloat)
107	* .toArray
108	* bufferedSource.close
109	*
110	* val processedAudioFloats = Seq(rawFloats).toDF("audio_content")
111	*
112	* val result = pipeline.fit(processedAudioFloats).transform(processedAudioFloats)
113	* result.select("text.result").show(truncate = false)
114	* +------------------------------------------------------------------------------------------+
115	* \|result \|
116	* +------------------------------------------------------------------------------------------+
117	* \|[MISTER QUILTER IS THE APOSTLE OF THE MIDLE CLASES AND WE ARE GLAD TO WELCOME HIS GOSPEL ]\|
118	* +------------------------------------------------------------------------------------------+
119	* }}}
120	*
121	* @param uid
122	* required uid for storing annotator to disk
123	* @groupname anno Annotator types
124	* @groupdesc anno
125	* Required input and expected output annotator types
126	* @groupname Ungrouped Members
127	* @groupname param Parameters
128	* @groupname setParam Parameter setters
129	* @groupname getParam Parameter getters
130	* @groupname Ungrouped Members
131	* @groupprio param 1
132	* @groupprio anno 2
133	* @groupprio Ungrouped 3
134	* @groupprio setParam 4
135	* @groupprio getParam 5
136	* @groupdesc param
137	* A list of (hyper-)parameter keys this annotator can take. Users can set and get the
138	* parameter values through setters and getters, respectively.
139	*/
140	class HubertForCTC(override val uid: String) extends Wav2Vec2ForCTC(uid) {
141
142	/** Annotator reference id. Used to identify elements in metadata or to refer to this annotator
143	* type
144	*/
145	def this() = this(Identifiable.randomUID("HubertForCTC"))	×
146
147	override def onWrite(path: String, spark: SparkSession): Unit = {
148	super.onWrite(path, spark)	×
149	writeTensorflowModelV2(	×
150	path,
151	spark,
152	getModelIfNotSet.tensorflowWrapper,	×
153	"_hubert_ctc",	×
154	HubertForCTC.tfFile,	×
155	configProtoBytes = getConfigProtoBytes)	×
156	}
157
158	}
159
160	trait ReadablePretrainedHubertForAudioModel
161	extends ParamsAndFeaturesReadable[HubertForCTC]
162	with HasPretrained[HubertForCTC] {
163	override val defaultModelName: Some[String] = Some("asr_hubert_large_ls960")	×
164
165	/** Java compliant-overrides */
166	override def pretrained(): HubertForCTC = super.pretrained()	×
167
168	override def pretrained(name: String): HubertForCTC = super.pretrained(name)	×
169
170	override def pretrained(name: String, lang: String): HubertForCTC =
171	super.pretrained(name, lang)	×
172
173	override def pretrained(name: String, lang: String, remoteLoc: String): HubertForCTC =
174	super.pretrained(name, lang, remoteLoc)	×
175	}
176
177	trait ReadHubertForAudioDLModel extends ReadTensorflowModel {
178	this: ParamsAndFeaturesReadable[HubertForCTC] =>
179
180	override val tfFile: String = "hubert_ctc_tensorflow"	×
181
182	def readTensorflow(instance: HubertForCTC, path: String, spark: SparkSession): Unit = {
183
184	val tf = readTensorflowModel(path, spark, "_hubert_ctc_tf", initAllTables = false)	×
185	instance.setModelIfNotSet(spark, tf)	×
186	}
187
188	addReader(readTensorflow)	×
189
190	def loadSavedModel(modelPath: String, spark: SparkSession): HubertForCTC = {
191
192	val (localModelPath, detectedEngine) = modelSanityCheck(modelPath)	×
193
194	val vocabJsonContent = loadJsonStringAsset(localModelPath, "vocab.json")	×
195	val vocabJsonMap =
196	parse(vocabJsonContent, useBigIntForLong = true).values	×
197	.asInstanceOf[Map[String, BigInt]]	×
198
199	val preprocessorConfigJsonContent =
200	loadJsonStringAsset(localModelPath, "preprocessor_config.json")	×
201	val preprocessorConfig =
202	Preprocessor.loadPreprocessorConfig(preprocessorConfigJsonContent)	×
203
204	/Universal parameters for all engines/
205	val annotatorModel = new HubertForCTC()
206	.setVocabulary(vocabJsonMap)
207	.setDoNormalize(preprocessorConfig.do_normalize)	×
208	.setFeatureSize(preprocessorConfig.feature_size)	×
209	.setPaddingSide(preprocessorConfig.padding_side)	×
210	.setPaddingValue(preprocessorConfig.padding_value)	×
211	.setReturnAttentionMask(preprocessorConfig.return_attention_mask)	×
212	.setSamplingRate(preprocessorConfig.sampling_rate)	×
213
214	annotatorModel.set(annotatorModel.engine, detectedEngine)	×
215
216	detectedEngine match {
217	case ModelEngine.tensorflow =>
218	val (wrapper, signatures) =	×
219	TensorflowWrapper.read(localModelPath, zipped = false, useBundle = true)
220
221	val _signatures = signatures match {
222	case Some(s) => s
223	case None => throw new Exception("Cannot load signature definitions from model!")	×
224	}
225
226	/** the order of setSignatures is important if we use getSignatures inside
227	* setModelIfNotSet
228	*/
229	annotatorModel
230	.setSignatures(_signatures)
231	.setModelIfNotSet(spark, wrapper)	×
232
233	case _ =>
234	throw new Exception(notSupportedEngineError)	×
235	}
236
237	annotatorModel
238	}
239	}
240
241	/** This is the companion object of [[HubertForCTC]]. Please refer to that class for the
242	* documentation.
243	*/
244	object HubertForCTC extends ReadablePretrainedHubertForAudioModel with ReadHubertForAudioDLModel

JohnSnowLabs / spark-nlp / 4413868535

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous