Update a database

A feature that is especially useful for databases growing over time, is the possibility to update a audformat.Database. For instance, consider the following database that contains age labels for a hundred files.

import audformat
import audformat.testing


db = audformat.testing.create_db(minimal=True)
db.schemes["age"] = audformat.Scheme(
    audformat.define.DataType.INTEGER,
    minimum=20,
    maximum=50,
)
audformat.testing.add_table(
    db,
    table_id="table",
    index_type=audformat.define.IndexType.FILEWISE,
    columns="age",
    num_files=100,
)
db

name: unittest
source: internal
usage: unrestricted
languages: [deu, eng]
schemes:
  age: {dtype: int, minimum: 20, maximum: 50}
tables:
  table:
    type: filewise
    columns:
      age: {scheme_id: age}

db["table"].df

	age
file
audio/001.wav	27
audio/002.wav	41
audio/003.wav	42
audio/004.wav	22
audio/005.wav	35
...	...
audio/096.wav	36
audio/097.wav	23
audio/098.wav	32
audio/099.wav	33
audio/100.wav	20

100 rows × 1 columns

Now assume we record more files to add to our original database. The new files are stored together with annotations in a second database, that is then added to the original database.

db_update = audformat.testing.create_db(minimal=True)
db_update.schemes["age"] = db.schemes["age"]
audformat.testing.add_table(
    db_update,
    table_id="table",
    index_type=audformat.define.IndexType.FILEWISE,
    columns="age",
    num_files=range(101, 105),
)
db.update(db_update)  # update original database with new data
db["table"].df

	age
file
audio/001.wav	27
audio/002.wav	41
audio/003.wav	42
audio/004.wav	22
audio/005.wav	35
...	...
audio/100.wav	20
audio/101.wav	38
audio/102.wav	23
audio/103.wav	48
audio/104.wav	26

104 rows × 1 columns

Or we find out that some files in the original database have wrong labels. To update those, we again start from a fresh database containing only the critical files, relabel them and then update the original database.

db_update = audformat.testing.create_db(minimal=True)
db_update.schemes["age"] = db.schemes["age"]
audformat.testing.add_table(
    db_update,
    table_id="table",
    index_type=audformat.define.IndexType.FILEWISE,
    columns="age",
    num_files=10,
)
db.update(db_update, overwrite=True)  # overwrite existing labels
db["table"].df

	age
file
audio/001.wav	27
audio/002.wav	31
audio/003.wav	31
audio/004.wav	45
audio/005.wav	24
...	...
audio/100.wav	20
audio/101.wav	38
audio/102.wav	23
audio/103.wav	48
audio/104.wav	26

104 rows × 1 columns

Finally, we want to add gender information to the database. Again, it might be easier to start with a fresh database to collect the new labels and only later merge it into our original database.

db_update = audformat.Database(
    name="update",
    languages=audformat.utils.map_language("french"),
)
db_update.schemes["gender"] = audformat.Scheme(
        labels=["female", "male"],
    )
audformat.testing.add_table(
    db_update,
    table_id="table",
    index_type=audformat.define.IndexType.FILEWISE,
    columns="gender",
    num_files=len(db.files),
)
db.update(db_update)
db["table"].df

	age	gender
file
audio/001.wav	27	female
audio/002.wav	31	male
audio/003.wav	31	female
audio/004.wav	45	female
audio/005.wav	24	male
...	...	...
audio/100.wav	20	male
audio/101.wav	38	male
audio/102.wav	23	male
audio/103.wav	48	female
audio/104.wav	26	male

104 rows × 2 columns

Note that this not only updates the table data, but also adds the new gender scheme:

db.schemes

age:
  {dtype: int, minimum: 20, maximum: 50}
gender:
  dtype: str
  labels: [female, male]