small fixes

README.md

@@ -29,7 +29,7 @@ configs:
 
 ## Overview
 
-BEANS-Zero is a **bioacoustics** benchmark designed to evaluate multimodal audio-language models in zero-shot settings. Introduced in the paper NatureLM-audio paper ([Robinson et al., 2024](https://openreview.net/forum?id=hJVdwBpWjt)), it brings together tasks from both existing datasets and newly curated resources.
+BEANS-Zero is a **bioacoustics** benchmark designed to evaluate multimodal audio-language models in zero-shot settings. Introduced in the paper NatureLM-audio paper ([Robinson et al., 2025](https://openreview.net/forum?id=hJVdwBpWjt)), it brings together tasks from both existing datasets and newly curated resources.
 
 The benchmark focuses on models that take a bioacoustic audio input (e.g., bird or mammal vocalizations) and a text instruction (e.g., "What species is in this audio?"), and return a textual output (e.g., "Taeniopygia guttata"). As a zero-shot benchmark, BEANS-Zero contains only a test split—no training or in-context examples are provided.
 
@@ -65,7 +65,7 @@ BEANS-Zero combines data from several well-known sources. There are total of 91,
 - `lifestage`: Predicting the lifestage of birds across multiple species (e.g., adult, juvenile), curated from [xeno-canto](https://xeno-canto.org/)
 - `call-type`: Classifying song vs. call across multiple bird species, curated from [xeno-canto](https://xeno-canto.org/)
 - `captioning`: Captioning bioacoustic audio on AnimalSpeak ([Robinson et al., 2024](https://doi.org/10.1109/ICASSP48485.2024.10447250))
-- `zf-indv`: Determining whether a recording contains multiplezebra finches, using programmatically generated mixtures (1–4 individuals) ([Elie and Theunissen, 2020](https://doi.org/10.6084/m9.figshare.11905533.v1))
+- `zf-indv`: Determining whether a recording contains multiplezebra finches, using programmatically generated mixtures (1–4 individuals) ([Elie and Theunissen, 2016](https://doi.org/10.6084/m9.figshare.11905533.v1))
 
 Each sample is labeled with its source dataset and license.
 
@@ -80,13 +80,13 @@ ds = load_dataset("EarthSpeciesProject/BEANS-Zero", split="test")
 print(ds)
 ```
 ```python
-# get audio for a sample
+# get audio for the first sample in the dataset, the 0th index
 audio = np.array(ds[0]["audio"])
 print(audio.shape)
 
 # get the instruction (prompt / query) for that sample
 print(ds[0]["instruction_text"])
-# the desired output (*only* used for evaluate)
+# the desired output (should *only* be used for evaluation)
 print(ds[0]["output"])
 
 # the component datasets of BEANS-Zero are:
@@ -113,15 +113,15 @@ print(sample.keys())
 The following fields are present in each example:
 
 - **source_dataset** (str): One of the source datasets mentioned above
-- **audio** (Sequence[float]): The audio data in float32 format. The audio is not decoded.
+- **audio** (Sequence[float]): The audio data as a list of floats.
 - **id** (str): Sample uuid.
 - **created_at** (str): Sample creation datetime in UTC
-- **metadata** (str): Each sample can have a different duration and a different sample rate. `sample_rate = json.loads(sample["metadata"])["sample_rate"]`
+- **metadata** (str): Each sample can have a different duration (in seconds) and a different sample rate (in Hz). The 'metadata' is a JSON string containing these two fields.
 - **file_name** (str): Audio file name
 - **instruction** (str): A prompt (a query) corresponding to the audio for your audio-text model with a placeholder for audio tokens. E.g. '<Audio><AudioHere></Audio> What is the scientific name for the focal species in the audio?'
 - **instruction_text** (str): Same as **instruction** but without the placeholder for audio tokens.
 - **output** (str): The expected output from the model
-- **task** (str): The task type e.g. classification / detection / pitch estimation / captioning.
+- **task** (str): The task type e.g. classification / detection / captioning.
 - **dataset_name** (str): Names corresponding to the evaluation tasks, e.g. 'esc50' or 'unseen-family-sci'.
 - **license** (str): The license of the dataset. For example, 'CC-BY-NC' or 'CC0'.