evaluation

README.md

@@ -8,7 +8,21 @@ datasets:
 - mozilla-foundation/common_voice_7_0
 model-index:
 - name: wav2vec2-xls-r-gn-cv7
-  results: []
+  results:
+  - task: 
+      name: Automatic Speech Recognition 
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice 7
+      type: mozilla-foundation/common_voice_7_0
+      args: pt
+    metrics:
+       - name: Validation WER
+         type: wer
+         value: 73.02
+       - name: Validation CER
+         type: cer
+         value: 17.79
 ---
 
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You

eval.py

@@ -0,0 +1,137 @@
+#!/usr/bin/env python3
+import argparse
+import re
+from typing import Dict
+
+import torch
+from datasets import Audio, Dataset, load_dataset, load_metric
+
+from transformers import AutoFeatureExtractor, pipeline
+
+
+def log_results(result: Dataset, args: Dict[str, str]):
+    """DO NOT CHANGE. This function computes and logs the result metrics."""
+
+    log_outputs = args.log_outputs
+    dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])
+
+    # load metric
+    wer = load_metric("wer")
+    cer = load_metric("cer")
+
+    # compute metrics
+    wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
+    cer_result = cer.compute(references=result["target"], predictions=result["prediction"])
+
+    # print & log results
+    result_str = f"WER: {wer_result}\n" f"CER: {cer_result}"
+    print(result_str)
+
+    with open(f"{dataset_id}_eval_results.txt", "w") as f:
+        f.write(result_str)
+
+    # log all results in text file. Possibly interesting for analysis
+    if log_outputs is not None:
+        pred_file = f"log_{dataset_id}_predictions.txt"
+        target_file = f"log_{dataset_id}_targets.txt"
+
+        with open(pred_file, "w") as p, open(target_file, "w") as t:
+
+            # mapping function to write output
+            def write_to_file(batch, i):
+                p.write(f"{i}" + "\n")
+                p.write(batch["prediction"] + "\n")
+                t.write(f"{i}" + "\n")
+                t.write(batch["target"] + "\n")
+
+            result.map(write_to_file, with_indices=True)
+
+
+def normalize_text(text: str) -> str:
+    """DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""
+
+    chars_to_ignore_regex = '[,?.!\-\;\:"“%‘”�—’…–]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+
+    text = re.sub(chars_to_ignore_regex, "", text.lower())
+
+    # In addition, we can normalize the target text, e.g. removing new lines characters etc...
+    # note that order is important here!
+    token_sequences_to_ignore = ["\n\n", "\n", "   ", "  "]
+
+    for t in token_sequences_to_ignore:
+        text = " ".join(text.split(t))
+
+    return text
+
+
+def main(args):
+    # load dataset
+    dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
+
+    # for testing: only process the first two examples as a test
+    # dataset = dataset.select(range(10))
+
+    # load processor
+    feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
+    sampling_rate = feature_extractor.sampling_rate
+
+    # resample audio
+    dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
+
+    # load eval pipeline
+    if args.device is None:
+        args.device = 0 if torch.cuda.is_available() else -1
+    asr = pipeline("automatic-speech-recognition", model=args.model_id, device=args.device)
+
+    # map function to decode audio
+    def map_to_pred(batch):
+        prediction = asr(
+            batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s
+        )
+
+        batch["prediction"] = prediction["text"]
+        batch["target"] = normalize_text(batch["sentence"])
+        return batch
+
+    # run inference on all examples
+    result = dataset.map(map_to_pred, remove_columns=dataset.column_names)
+
+    # compute and log_results
+    # do not change function below
+    log_results(result, args)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers"
+    )
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        required=True,
+        help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets",
+    )
+    parser.add_argument(
+        "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'`  for Common Voice"
+    )
+    parser.add_argument("--split", type=str, required=True, help="Split of the dataset. *E.g.* `'test'`")
+    parser.add_argument(
+        "--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to 5 seconds."
+    )
+    parser.add_argument(
+        "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to 1 second."
+    )
+    parser.add_argument(
+        "--log_outputs", action="store_true", help="If defined, write outputs to log file for analysis."
+    )
+    parser.add_argument(
+        "--device",
+        type=int,
+        default=None,
+        help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.",
+    )
+    args = parser.parse_args()
+
+    main(args)

eval.sh

@@ -0,0 +1 @@
+./eval.py --model_id lgris/wav2vec2-xls-r-gn-cv7 --dataset mozilla-foundation/common_voice_7_0 --config gn --split validation --log_outputs

log_mozilla-foundation_common_voice_7_0_gn_validation_predictions.txt

@@ -0,0 +1,72 @@
+0
+nykotererei ho oysã porã
+1
+mávapape mba'e ikuakapáva poyvórape
+2
+ikatuhá peve ahejande chupe
+3
+povákuri iperiumguekuéra ndive tava parauguavýpe
+4
+kui mba'e oiku vapynandiovasry ha oñemombeporaguáiva
+5
+ohopáma piko
+6
+nomberekyra
+7
+he rembiapokue kuy
+8
+ñe'epyrũrãme ndahe raivaʼekue
+9
+ha upépe oĩjepe oheiakaʼe me lipiráriepe
+10
+romaña porãva ojuhe noroñe'ẽi ojupe
+11
+na chemandu'ái
+12
+hoyhepyme'ẽ chupe
+13
+oipurúrheikuái ka'a ha ho'ute rere
+14
+oré rotuviatã lunekue
+15
+mbapéipate hendume umi rugárpe
+16
+mávapape mba'e ñanembopyʼáruribéga
+17
+omosã umi mbaʼekuaa
+18
+po'ã ka hape aréko heta mba'etuja ndaipotãvéi uma
+19
+péia ha'e
+20
+mba'e riko chépe
+21
+aikéma katu porapýpe
+22
+nápepe oĩ cherajyla pinsésa
+23
+ohasahárupi a'ãraña voo karai purvensio
+24
+aoñetépaʼere
+25
+tata guasu opu'ãvo 'ojahéi tekoháre
+26
+upéima niko ñelmo'ãku oñeñarundujevýva
+27
+ohecha mba'éichapa ha'e oñemboja ha oike guasura'y mbive yro'õme
+28
+ño nemborieahúko na ñandiapo'ãivavoi
+29
+ha oporeíma hesy
+30
+oopa umi mba'ére niko tapichakuéra ohaihúi chupe
+31
+ndaipóri ivy pome'ẽ'aréva hi'a
+32
+aravo opa jeike hagua
+33
+oimérez nriko hovah hína plá ta yguy mba'e
+34
+a upéicha avei ndoipotãete uñe'ẽkui mba'ekuéra ndive
+35
+huungy ha morote

log_mozilla-foundation_common_voice_7_0_gn_validation_targets.txt

@@ -0,0 +1,72 @@
+0
+mmm ko terere hoysã porã
+1
+mávapa pe mbaʼe ipuʼakapáva ko yvórape
+2
+ikatuha peve ahejánte chupe
+3
+ovákuri ipehẽnguekuéra ndive táva paraguaýpe
+4
+kuimba'e oikóva pynandi hovasy ha oñemonde paraguáiva
+5
+ohopámapiko
+6
+ndekyra
+7
+hembiapokue ypy
+8
+ñepyrũrãme ndaheraivaʼekue
+9
+ha upépe oĩjepe avei academia literariape
+10
+romaña porã ojuehe noroñe'ẽi ojupe
+11
+nachemandu'ái
+12
+ohepyme'ẽ chupe
+13
+oipuru hikuái ka'a ha ho'u terere
+14
+ore rostudiáta luneskue
+15
+mbaʼépa pehendúne umi lugarpe
+16
+mávapa pe mbaʼe ñanembopyʼaroryvéva
+17
+omosã umi mbaʼekuaa
+18
+ko'ã cajape areko heta mba'e tuja ndaipotavéima
+19
+péa ha'e
+20
+mba'épiko chéve
+21
+aikéma katu korapýpe
+22
+napépe oĩ che rajy la princesa
+23
+ohasaha rupi ára ñavõ karai prudencio
+24
+añetépa ere
+25
+tata guasu opu'ãvo ojahéi tekoháre
+26
+upéima niko ñaimo'ã ku añeñandujeýva
+27
+ohecha mba'éichapa ha'e oñemboja ha oike guasu ra'y ndive yno'õme
+28
+ñande mboriahúko nañandepo'aivavoi
+29
+ha ohoreíma heseve
+30
+opa umi mba'ére niko tapichakuéra ohayhu ichupe
+31
+ndaipóri yvy ome'ẽ'aréva hi'a
+32
+aravo ojeike haguã
+33
+oiméne niko kóva hína pláta yvyguy mba'e
+34
+ha upéicha avei ndoipotaiete oñe'ẽ kuimba'ekuéra ndive
+35
+hũngy ha morotĩ

mozilla-foundation_common_voice_7_0_gn_validation_eval_results.txt

@@ -0,0 +1,2 @@
+WER: 0.7302631578947368
+CER: 0.17799043062200956