Commit
·
f73dbf6
1
Parent(s):
1e4af7c
up
Browse files
app.py
CHANGED
|
@@ -14,10 +14,24 @@ from huggingface_hub import Repository
|
|
| 14 |
REFERENCE_NAME = "references"
|
| 15 |
SUBMISSION_NAME = "submissions"
|
| 16 |
|
| 17 |
-
REFERENCE_URL = os.path.join(
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
-
TEST_SETS = [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
EXPECTED_TEST_FILES = [f + ".txt" for f in TEST_SETS]
|
| 22 |
OPTIONAL_TEST_SETS = ["switch-board", "call-home", "chime-4"]
|
| 23 |
|
|
@@ -28,7 +42,9 @@ HF_TOKEN = os.environ.get("HF_TOKEN")
|
|
| 28 |
|
| 29 |
|
| 30 |
def compute_wer(pred_file, ref_file):
|
| 31 |
-
with open(pred_file, "r", encoding="utf-8") as pred, open(
|
|
|
|
|
|
|
| 32 |
pred_lines = [line.strip() for line in pred.readlines()]
|
| 33 |
ref_lines = [line.strip() for line in ref.readlines()]
|
| 34 |
|
|
@@ -36,11 +52,19 @@ def compute_wer(pred_file, ref_file):
|
|
| 36 |
return wer
|
| 37 |
|
| 38 |
|
| 39 |
-
reference_repo = Repository(
|
| 40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
submission_repo.git_pull()
|
| 42 |
|
| 43 |
-
all_submissions = [
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
|
| 45 |
all_results = read_csv(CSV_RESULTS_FILE)
|
| 46 |
evaluated_submissions = all_results["name"].values.tolist()
|
|
@@ -55,7 +79,9 @@ if len(non_evaluated_submissions) > 0:
|
|
| 55 |
submitted_files = [f for f in submitted_files if f in EXPECTED_TEST_FILES]
|
| 56 |
|
| 57 |
if sorted(EXPECTED_TEST_FILES) != sorted(submitted_files):
|
| 58 |
-
raise ValueError(
|
|
|
|
|
|
|
| 59 |
|
| 60 |
for file in submitted_files:
|
| 61 |
ref_file = os.path.join(REFERENCE_NAME, file)
|
|
@@ -76,7 +102,19 @@ if len(non_evaluated_submissions) > 0:
|
|
| 76 |
commit_url = reference_repo.push_to_hub()
|
| 77 |
print(commit_url)
|
| 78 |
|
| 79 |
-
COLUMN_NAMES = {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
|
| 81 |
table = all_results.round(4)
|
| 82 |
table = table.rename(columns=COLUMN_NAMES)
|
|
@@ -110,7 +148,11 @@ st.markdown("To submit to ESC, please click on the instructions below ↓")
|
|
| 110 |
|
| 111 |
st.markdown("TODO: Add instructions ...")
|
| 112 |
|
| 113 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
|
| 115 |
-
if
|
| 116 |
-
st.write(
|
|
|
|
| 14 |
REFERENCE_NAME = "references"
|
| 15 |
SUBMISSION_NAME = "submissions"
|
| 16 |
|
| 17 |
+
REFERENCE_URL = os.path.join(
|
| 18 |
+
"https://huggingface.co/datasets/esc-bench", REFERENCE_NAME
|
| 19 |
+
)
|
| 20 |
+
SUBMISSION_URL = os.path.join(
|
| 21 |
+
"https://huggingface.co/datasets/esc-bench", SUBMISSION_NAME
|
| 22 |
+
)
|
| 23 |
|
| 24 |
+
TEST_SETS = [
|
| 25 |
+
"librispeech-clean",
|
| 26 |
+
"librispeech-other",
|
| 27 |
+
"common-voice-9",
|
| 28 |
+
"vox-populi",
|
| 29 |
+
"ted-lium",
|
| 30 |
+
"giga-speech",
|
| 31 |
+
"spgi-speech",
|
| 32 |
+
"earnings-22",
|
| 33 |
+
"ami",
|
| 34 |
+
]
|
| 35 |
EXPECTED_TEST_FILES = [f + ".txt" for f in TEST_SETS]
|
| 36 |
OPTIONAL_TEST_SETS = ["switch-board", "call-home", "chime-4"]
|
| 37 |
|
|
|
|
| 42 |
|
| 43 |
|
| 44 |
def compute_wer(pred_file, ref_file):
|
| 45 |
+
with open(pred_file, "r", encoding="utf-8") as pred, open(
|
| 46 |
+
ref_file, "r", encoding="utf-8"
|
| 47 |
+
) as ref:
|
| 48 |
pred_lines = [line.strip() for line in pred.readlines()]
|
| 49 |
ref_lines = [line.strip() for line in ref.readlines()]
|
| 50 |
|
|
|
|
| 52 |
return wer
|
| 53 |
|
| 54 |
|
| 55 |
+
reference_repo = Repository(
|
| 56 |
+
local_dir="references", clone_from=REFERENCE_URL, use_auth_token=HF_TOKEN
|
| 57 |
+
)
|
| 58 |
+
submission_repo = Repository(
|
| 59 |
+
local_dir="submissions", clone_from=SUBMISSION_URL, use_auth_token=HF_TOKEN
|
| 60 |
+
)
|
| 61 |
submission_repo.git_pull()
|
| 62 |
|
| 63 |
+
all_submissions = [
|
| 64 |
+
folder
|
| 65 |
+
for folder in os.listdir(SUBMISSION_NAME)
|
| 66 |
+
if os.path.isdir(os.path.join(SUBMISSION_NAME, folder)) and folder != ".git"
|
| 67 |
+
]
|
| 68 |
|
| 69 |
all_results = read_csv(CSV_RESULTS_FILE)
|
| 70 |
evaluated_submissions = all_results["name"].values.tolist()
|
|
|
|
| 79 |
submitted_files = [f for f in submitted_files if f in EXPECTED_TEST_FILES]
|
| 80 |
|
| 81 |
if sorted(EXPECTED_TEST_FILES) != sorted(submitted_files):
|
| 82 |
+
raise ValueError(
|
| 83 |
+
f"{', '.join(submitted_files)} were submitted, but expected {', '.join(EXPECTED_TEST_FILES)}"
|
| 84 |
+
)
|
| 85 |
|
| 86 |
for file in submitted_files:
|
| 87 |
ref_file = os.path.join(REFERENCE_NAME, file)
|
|
|
|
| 102 |
commit_url = reference_repo.push_to_hub()
|
| 103 |
print(commit_url)
|
| 104 |
|
| 105 |
+
COLUMN_NAMES = {
|
| 106 |
+
"librispeech-clean": "ls-clean",
|
| 107 |
+
"librispeech-other": "ls-other",
|
| 108 |
+
"common-voice-9": "cv9",
|
| 109 |
+
"vox-populi": "vox",
|
| 110 |
+
"ted-lium": "ted",
|
| 111 |
+
"giga-speech": "giga",
|
| 112 |
+
"spgi-speech": "spgi",
|
| 113 |
+
"earnings-22": "e22",
|
| 114 |
+
"ami": "ami",
|
| 115 |
+
"chime-4": "chime",
|
| 116 |
+
"switch-board": "swbd",
|
| 117 |
+
}
|
| 118 |
|
| 119 |
table = all_results.round(4)
|
| 120 |
table = table.rename(columns=COLUMN_NAMES)
|
|
|
|
| 148 |
|
| 149 |
st.markdown("TODO: Add instructions ...")
|
| 150 |
|
| 151 |
+
# Using the "with" syntax
|
| 152 |
+
with st.form(key="my_form"):
|
| 153 |
+
text_input = st.text_input(label="Name")
|
| 154 |
+
uploaded_file = st.file_uploader("Choose a zip file")
|
| 155 |
+
submit_button = st.form_submit_button(label="Submit")
|
| 156 |
|
| 157 |
+
if submit_button:
|
| 158 |
+
st.write("Computing scores ...")
|